chore: sync code and project files

Dockerfile

@@ -7,6 +7,7 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
     libsm6 \
     libxext6 \
     libgl1 \
+    librsvg2-bin \
     fonts-noto-cjk \
     fonts-wqy-zenhei \
     curl \

api/celery_app.py

@@ -0,0 +1,85 @@
+"""
+Celery 应用配置
+支持异步任务处理，可水平扩展 Worker
+"""
+import os
+import sys
+from pathlib import Path
+
+# 确保项目根目录在 path 中
+PROJECT_ROOT = Path(__file__).parent.parent
+sys.path.insert(0, str(PROJECT_ROOT))
+
+from celery import Celery
+
+# Redis 配置 (可通过环境变量覆盖)
+REDIS_URL = os.getenv("REDIS_URL", "redis://localhost:6379/0")
+
+# 创建 Celery 应用
+celery_app = Celery(
+    "video_flow",
+    broker=REDIS_URL,
+    backend=REDIS_URL,
+    include=["api.tasks.video_tasks", "api.tasks.audio_tasks"]
+)
+
+# Celery 配置
+celery_app.conf.update(
+    # 任务序列化
+    task_serializer="json",
+    accept_content=["json"],
+    result_serializer="json",
+    
+    # 时区
+    timezone="Asia/Shanghai",
+    enable_utc=True,
+    
+    # 任务追踪
+    task_track_started=True,
+    task_time_limit=600,  # 10分钟超时
+    task_soft_time_limit=540,  # 9分钟软超时
+    
+    # 结果保存
+    result_expires=3600,  # 1小时后过期
+    
+    # Worker 配置
+    worker_prefetch_multiplier=1,  # 每次只取一个任务，适合长任务
+    worker_concurrency=2,  # 每个 Worker 的并发数
+    
+    # 任务路由 (可选，用于任务分类)
+    task_routes={
+        "api.tasks.video_tasks.*": {"queue": "video"},
+        "api.tasks.audio_tasks.*": {"queue": "audio"},
+    },
+    
+    # 默认队列
+    task_default_queue="default",
+)
+
+# 健康检查任务
+@celery_app.task(bind=True)
+def health_check(self):
+    """Worker 健康检查"""
+    return {
+        "status": "ok",
+        "worker_id": self.request.id,
+        "hostname": self.request.hostname
+    }
+
+
+if __name__ == "__main__":
+    celery_app.start()
+
+
+
+
+
+
+
+
+
+
+
+
+
+

api/main.py

@@ -68,6 +68,7 @@ app.add_middleware(
 app.mount("/static/output", StaticFiles(directory=str(config.OUTPUT_DIR)), name="output")
 app.mount("/static/temp", StaticFiles(directory=str(config.TEMP_DIR)), name="temp")
 app.mount("/static/assets", StaticFiles(directory=str(config.ASSETS_DIR)), name="assets")
+app.mount("/static/fonts", StaticFiles(directory=str(config.FONTS_DIR)), name="fonts")
 
 # Legacy mounts（8502 runtime 产物，宿主机目录通过 docker-compose 挂载到容器内）
 try:

api/routes/__init__.py

@@ -0,0 +1,15 @@
+# API Routes Package
+
+
+
+
+
+
+
+
+
+
+
+
+
+

api/routes/assets.py

@@ -0,0 +1,408 @@
+"""
+素材管理 API 路由
+提供素材上传、列表、删除功能
+"""
+import os
+import time
+import logging
+from typing import List, Optional
+from pathlib import Path
+
+from fastapi import APIRouter, HTTPException, UploadFile, File
+from fastapi.responses import FileResponse, RedirectResponse
+from pydantic import BaseModel
+
+import config
+from modules.db_manager import db
+from modules.legacy_path_mapper import map_legacy_local_path
+
+logger = logging.getLogger(__name__)
+router = APIRouter()
+
+
+# ============================================================
+# Pydantic Models
+# ============================================================
+
+class AssetInfo(BaseModel):
+    id: int
+    project_id: str
+    scene_id: int
+    asset_type: str
+    status: str
+    local_path: Optional[str] = None
+    url: Optional[str] = None
+    metadata: dict = {}
+
+
+class UploadResponse(BaseModel):
+    success: bool
+    filename: str
+    path: str
+    url: str
+
+
+class StickerItem(BaseModel):
+    id: str
+    name: str
+    url: str
+    kind: str = "builtin"  # builtin/custom
+    tags: List[str] = []
+    category: Optional[str] = None
+    license: Optional[str] = None
+    attribution: Optional[str] = None
+
+
+# ============================================================
+# API Endpoints
+# ============================================================
+
+@router.get("/list/{project_id}", response_model=List[AssetInfo])
+async def list_assets(
+    project_id: str,
+    asset_type: Optional[str] = None
+):
+    """列出项目的所有素材"""
+    assets = db.get_assets(project_id, asset_type)
+    
+    result = []
+    for asset in assets:
+        # 添加 URL
+        url = None
+        local_path = asset.get("local_path")
+        remote_url = asset.get("remote_url")
+
+        # 1) 直接可见路径：走 file proxy
+        if local_path and os.path.exists(local_path):
+            url = f"/api/assets/file/{asset.get('id', 0)}"
+        # 1.5) 远端 URL（例如 R2/CDN）。本地缺失时也要能预览
+        elif remote_url and isinstance(remote_url, str) and remote_url.strip():
+            url = remote_url.strip()
+        else:
+            # 2) legacy 映射（/root/video-flow -> /legacy/* 或 /app/*）
+            mapped_path, mapped_url = map_legacy_local_path(local_path)
+            if mapped_path and os.path.exists(mapped_path):
+                # 静态 URL 不一定覆盖子目录；优先静态（若提供），否则走 file proxy
+                url = mapped_url or f"/api/assets/file/{asset.get('id', 0)}"
+            else:
+                # 3) 兜底：某些历史数据用 asset_id 命名（例如 /legacy/temp/1.png）
+                aid = int(asset.get("id") or 0)
+                if aid > 0:
+                    # 优先按原扩展名尝试
+                    ext = Path(local_path).suffix.lower() if isinstance(local_path, str) else ""
+                    candidates = []
+                    if ext:
+                        candidates.append((f"/legacy/temp/{aid}{ext}", f"/static/legacy-temp/{aid}{ext}"))
+                    # 常见图片扩展
+                    for e in [".png", ".jpg", ".jpeg", ".webp"]:
+                        candidates.append((f"/legacy/temp/{aid}{e}", f"/static/legacy-temp/{aid}{e}"))
+                    for p, u in candidates:
+                        if os.path.exists(p):
+                            url = u
+                            break
+        
+        result.append(AssetInfo(
+            id=asset.get("id", 0),
+            project_id=asset["project_id"],
+            scene_id=asset["scene_id"],
+            asset_type=asset["asset_type"],
+            status=asset["status"],
+            local_path=asset.get("local_path"),
+            url=url,
+            metadata=asset.get("metadata", {})
+        ))
+    
+    return result
+
+
+@router.get("/file/{asset_id}")
+async def get_asset_file(asset_id: int):
+    """按 asset_id 直接返回文件（用于前端预览/拖拽素材）。"""
+    a = db.get_asset_by_id(int(asset_id))
+    if not a:
+        raise HTTPException(status_code=404, detail="素材不存在")
+    local_path = a.get("local_path")
+    remote_url = a.get("remote_url")
+
+    # 1) 原路径
+    if local_path and os.path.exists(local_path):
+        return FileResponse(path=local_path, filename=Path(local_path).name)
+
+    # 1.5) 远端 URL：本地缺失时直接重定向（避免前端 404 黑屏）
+    if remote_url and isinstance(remote_url, str) and remote_url.strip():
+        return RedirectResponse(url=remote_url.strip(), status_code=307)
+
+    # 2) legacy 映射
+    mapped_path, _ = map_legacy_local_path(local_path)
+    if mapped_path and os.path.exists(mapped_path):
+        return FileResponse(path=mapped_path, filename=Path(mapped_path).name)
+
+    # 3) asset_id 命名兜底（常见于历史图片：/legacy/temp/{id}.png）
+    aid = int(a.get("id") or 0)
+    ext = Path(local_path).suffix.lower() if isinstance(local_path, str) else ""
+    candidates = []
+    if aid > 0:
+        if ext:
+            candidates.append(f"/legacy/temp/{aid}{ext}")
+        for e in [".png", ".jpg", ".jpeg", ".webp", ".mp4", ".mov", ".m4a", ".mp3"]:
+            candidates.append(f"/legacy/temp/{aid}{e}")
+    for p in candidates:
+        if os.path.exists(p):
+            return FileResponse(path=p, filename=Path(p).name)
+
+    raise HTTPException(status_code=404, detail="素材文件不存在")
+
+
+@router.post("/upload", response_model=UploadResponse)
+async def upload_asset(
+    file: UploadFile = File(...),
+    asset_type: str = "custom"
+):
+    """
+    上传自定义素材
+    支持图片、视频、音频
+    """
+    # 验证文件类型
+    allowed_extensions = {
+        "image": [".jpg", ".jpeg", ".png", ".gif", ".webp"],
+        "video": [".mp4", ".mov", ".avi", ".mkv", ".webm"],
+        "audio": [".mp3", ".wav", ".m4a", ".aac"],
+        "custom": [".jpg", ".jpeg", ".png", ".gif", ".webp", 
+                   ".mp4", ".mov", ".avi", ".mkv", ".webm",
+                   ".mp3", ".wav", ".m4a", ".aac"]
+    }
+    
+    ext = Path(file.filename).suffix.lower()
+    if ext not in allowed_extensions.get(asset_type, allowed_extensions["custom"]):
+        raise HTTPException(
+            status_code=400, 
+            detail=f"不支持的文件类型: {ext}"
+        )
+    
+    # 生成唯一文件名
+    timestamp = int(time.time())
+    new_filename = f"{asset_type}_{timestamp}{ext}"
+    file_path = config.TEMP_DIR / new_filename
+    
+    # 保存文件
+    try:
+        content = await file.read()
+        with open(file_path, "wb") as f:
+            f.write(content)
+        
+        url = f"/static/temp/{new_filename}"
+        
+        return UploadResponse(
+            success=True,
+            filename=new_filename,
+            path=str(file_path),
+            url=url
+        )
+        
+    except Exception as e:
+        logger.error(f"文件上传失败: {e}")
+        raise HTTPException(status_code=500, detail=str(e))
+
+
+@router.get("/download/{filename}")
+async def download_asset(filename: str):
+    """下载素材文件"""
+    # 查找文件
+    for directory in [config.OUTPUT_DIR, config.TEMP_DIR]:
+        file_path = directory / filename
+        if file_path.exists():
+            return FileResponse(
+                path=str(file_path),
+                filename=filename,
+                media_type="application/octet-stream"
+            )
+    
+    raise HTTPException(status_code=404, detail="文件不存在")
+
+
+@router.delete("/{asset_id}")
+async def delete_asset(asset_id: int):
+    """删除素材"""
+    # TODO: 实现真正的删除逻辑
+    logger.info(f"删除素材: {asset_id}")
+    return {"message": "素材已删除", "id": asset_id}
+
+
+@router.get("/bgm")
+async def list_bgm():
+    """列出可用的 BGM"""
+    bgm_dir = config.ASSETS_DIR / "bgm"
+    if not bgm_dir.exists():
+        return []
+    
+    bgm_files = []
+    for f in bgm_dir.iterdir():
+        if f.suffix.lower() in ['.mp3', '.mp4', '.m4a', '.wav']:
+            bgm_files.append({
+                "id": f.name,
+                "name": f.stem,
+                "path": str(f),
+                "url": f"/static/assets/bgm/{f.name}"
+            })
+    
+    return bgm_files
+
+
+@router.get("/fonts")
+async def list_fonts():
+    """列出可用的字体"""
+    fonts = []
+    
+    # 项目字体
+    if config.FONTS_DIR.exists():
+        for f in config.FONTS_DIR.iterdir():
+            if f.suffix.lower() in ['.ttf', '.otf', '.ttc']:
+                fonts.append({
+                    "id": f.name,
+                    "name": f.stem,
+                    "path": str(f),
+                    "url": f"/static/fonts/{f.name}",
+                    # 前端预览可直接使用 id 作为 CSS font-family（配合 FontFace 动态加载）
+                    "css_family": f.name
+                })
+    
+    # 添加系统字体选项
+    fonts.append({
+        "id": "system-pingfang",
+        "name": "苹方 (系统)",
+        "path": "/System/Library/Fonts/PingFang.ttc",
+        "url": None,
+        "css_family": "PingFang SC"
+    })
+    
+    return fonts
+
+
+@router.get("/stickers")
+async def list_stickers():
+    """列出贴纸（PNG/SVG），用于左侧贴纸库。"""
+    builtin_dir = config.ASSETS_DIR / "stickers_builtin"
+    custom_dir = config.ASSETS_DIR / "stickers_custom"
+    custom_dir.mkdir(parents=True, exist_ok=True)
+
+    items: List[dict] = []
+
+    # builtin: 优先读 index.json（支持分类/标签/授权说明）
+    index_path = builtin_dir / "index.json"
+    if index_path.exists():
+        try:
+            import json
+            data = json.loads(index_path.read_text(encoding="utf-8"))
+            pack = data.get("pack") or {}
+            for cat in (data.get("categories") or []):
+                cat_name = cat.get("name") or cat.get("id") or "未分类"
+                for it in (cat.get("items") or []):
+                    f = str(it.get("file") or "")
+                    if not f:
+                        continue
+                    url = f"/static/assets/stickers_builtin/{f}"
+                    items.append({
+                        "id": str(it.get("id") or f),
+                        "name": str(it.get("name") or it.get("id") or f),
+                        "url": url,
+                        "kind": "builtin",
+                        "tags": it.get("tags") or [],
+                        "category": cat_name,
+                        "license": pack.get("license"),
+                        "attribution": pack.get("attribution"),
+                    })
+        except Exception as e:
+            logger.warning(f"Failed to read stickers_builtin/index.json: {e}")
+
+    # custom: 扫描目录
+    if custom_dir.exists():
+        for f in custom_dir.iterdir():
+            if f.suffix.lower() not in [".png", ".svg", ".webp"]:
+                continue
+            items.append({
+                "id": f"custom-{f.name}",
+                "name": f.stem,
+                "url": f"/static/assets/stickers_custom/{f.name}",
+                "kind": "custom",
+                "tags": [],
+                "category": "自定义",
+                "license": None,
+                "attribution": None,
+            })
+
+    return items
+
+
+@router.post("/stickers/upload")
+async def upload_sticker(file: UploadFile = File(...)):
+    """上传贴纸（png/svg/webp）到 assets/stickers_custom。"""
+    ext = Path(file.filename).suffix.lower()
+    if ext not in [".png", ".svg", ".webp"]:
+        raise HTTPException(status_code=400, detail=f"不支持的贴纸类型: {ext}")
+    target_dir = config.ASSETS_DIR / "stickers_custom"
+    target_dir.mkdir(parents=True, exist_ok=True)
+    safe_name = Path(file.filename).name
+    target = target_dir / safe_name
+    if target.exists():
+        ts = int(time.time())
+        target = target_dir / f"{Path(safe_name).stem}_{ts}{ext}"
+    content = await file.read()
+    with open(target, "wb") as f:
+        f.write(content)
+    return {
+        "success": True,
+        "id": f"custom-{target.name}",
+        "name": target.stem,
+        "url": f"/static/assets/stickers_custom/{target.name}",
+    }
+
+
+@router.post("/fonts/upload")
+async def upload_font(file: UploadFile = File(...)):
+    """上传字体（ttf/otf/ttc），保存到 assets/fonts，供字幕/花字使用。"""
+    ext = Path(file.filename).suffix.lower()
+    if ext not in [".ttf", ".otf", ".ttc"]:
+        raise HTTPException(status_code=400, detail=f"不支持的字体类型: {ext}")
+
+    config.FONTS_DIR.mkdir(parents=True, exist_ok=True)
+    safe_name = Path(file.filename).name
+    target = config.FONTS_DIR / safe_name
+
+    # 若同名存在，则加时间戳避免覆盖
+    if target.exists():
+        ts = int(time.time())
+        target = config.FONTS_DIR / f"{Path(safe_name).stem}_{ts}{ext}"
+
+    try:
+        content = await file.read()
+        if not content or len(content) < 1024:
+            raise HTTPException(status_code=400, detail="字体文件为空或损坏")
+        with open(target, "wb") as f:
+            f.write(content)
+        return {
+            "success": True,
+            "id": target.name,
+            "name": target.stem,
+            "path": str(target),
+            "url": f"/static/fonts/{target.name}",
+            "css_family": target.name,
+        }
+    except HTTPException:
+        raise
+    except Exception as e:
+        logger.error(f"字体上传失败: {e}")
+        raise HTTPException(status_code=500, detail=str(e))
+
+
+
+
+
+
+
+
+
+
+
+
+
+

api/routes/compose.py

@@ -0,0 +1,837 @@
+"""
+视频合成 API 路由
+基于编辑器状态进行最终合成
+支持同步/异步两种模式，异步模式通过 Celery 任务队列处理
+"""
+import os
+import time
+import logging
+import subprocess
+from typing import List, Optional, Dict, Any
+from pathlib import Path
+
+from fastapi import APIRouter, HTTPException, BackgroundTasks
+from pydantic import BaseModel
+
+import config
+from modules.db_manager import db
+from modules.composer import VideoComposer
+from modules import ffmpeg_utils, factory
+from modules.text_renderer import renderer
+
+# Celery 任务导入 (可选，默认关闭；需要显式开启 USE_CELERY=1)
+CELERY_AVAILABLE = False
+if str(os.getenv("USE_CELERY", "")).lower() in ("1", "true", "yes", "on"):
+    try:
+        from api.tasks.video_tasks import compose_from_script_task, compose_from_tracks_task
+        CELERY_AVAILABLE = True
+    except ImportError:
+        CELERY_AVAILABLE = False
+
+logger = logging.getLogger(__name__)
+router = APIRouter()
+
+
+# ============================================================
+# Pydantic Models
+# ============================================================
+
+class ComposeRequest(BaseModel):
+    """合成请求"""
+    project_id: str
+    
+    # 轨道数据（从编辑器状态转换）
+    video_clips: List[Dict[str, Any]] = []
+    voiceover_clips: List[Dict[str, Any]] = []
+    subtitle_clips: List[Dict[str, Any]] = []
+    fancy_text_clips: List[Dict[str, Any]] = []
+    sticker_clips: List[Dict[str, Any]] = []
+    bgm_clip: Optional[Dict[str, Any]] = None
+    
+    # 全局设置
+    voice_type: str = "zh_female_santongyongns_saturn_bigtts"
+    bgm_volume: float = 0.15
+    output_name: Optional[str] = None
+
+
+def _validate_no_overlap(clips: List[Dict[str, Any]], *, min_gap: float = 0.0) -> None:
+    """Validate clips are non-overlapping when sorted by start time."""
+    ordered = sorted(clips, key=lambda x: float(x.get("start") or 0))
+    prev_end = 0.0
+    for c in ordered:
+        start = float(c.get("start") or 0.0)
+        dur = float(c.get("duration") or 0.0)
+        end = start + dur
+        if start + 1e-6 < prev_end - min_gap:
+            raise ValueError("视频片段存在重叠，请先在时间轴调整避免同轨道覆盖。")
+        prev_end = max(prev_end, end)
+
+
+def _validate_trim_bounds(video_clips: List[Dict[str, Any]]) -> None:
+    """
+    Validate trim bounds are sane and, if source_duration is provided, do not exceed it.
+    """
+    for c in video_clips:
+        trim_start = float(c.get("trim_start") or 0.0)
+        trim_end = c.get("trim_end")
+        duration = float(c.get("duration") or 0.0)
+        if trim_end is None:
+            trim_end = trim_start + duration
+        trim_end = float(trim_end)
+        if trim_end <= trim_start + 1e-6:
+            raise ValueError("视频片段裁剪区间无效：trim_end 必须大于 trim_start。")
+
+        src_dur = c.get("source_duration")
+        if src_dur is not None:
+            try:
+                src_dur_f = float(src_dur)
+                if src_dur_f > 0 and trim_end > src_dur_f + 1e-3:
+                    raise ValueError("视频片段裁剪越界：trim_end 超过源视频时长，请先缩短片段或调整 trim。")
+            except ValueError:
+                raise
+            except Exception:
+                # ignore parsing errors
+                pass
+
+
+class ComposeResponse(BaseModel):
+    success: bool
+    message: str
+    output_path: Optional[str] = None
+    output_url: Optional[str] = None
+    task_id: Optional[str] = None
+
+
+class ComposeStatus(BaseModel):
+    status: str  # queued, pending, processing, completed, failed, cancelled
+    progress: float = 0
+    message: str = ""
+    output_path: Optional[str] = None
+    output_url: Optional[str] = None
+
+
+def _persist_job(task_id: str, project_id: str, *, status: str, progress: float, message: str, request: Optional[Dict[str, Any]] = None):
+    """Persist status to DB (render_jobs)."""
+    try:
+        existing = db.get_render_job(task_id)
+        if not existing:
+            db.create_render_job(task_id, project_id, status=status, progress=progress, message=message, request=request or {})
+        else:
+            db.update_render_job(task_id, {"status": status, "progress": progress, "message": message})
+    except Exception as e:
+        logger.debug(f"persist_job failed (non-fatal): {e}")
+
+
+# ============================================================
+# API Endpoints
+# ============================================================
+
+@router.post("/render", response_model=ComposeResponse)
+async def render_video(request: ComposeRequest, background_tasks: BackgroundTasks):
+    """
+    异步合成视频
+    优先使用 Celery 任务队列（支持水平扩展）
+    如果 Celery 不可用，降级为 FastAPI BackgroundTasks
+    """
+    project = db.get_project(request.project_id)
+    if not project:
+        raise HTTPException(status_code=404, detail="项目不存在")
+    
+    # 优先使用 Celery 任务队列
+    if CELERY_AVAILABLE:
+        try:
+            task = compose_from_tracks_task.delay(
+                project_id=request.project_id,
+                video_clips=request.video_clips,
+                voiceover_clips=request.voiceover_clips,
+                subtitle_clips=request.subtitle_clips,
+                fancy_text_clips=request.fancy_text_clips,
+                bgm_clip=request.bgm_clip,
+                voice_type=request.voice_type,
+                bgm_volume=request.bgm_volume,
+                output_name=request.output_name
+            )
+            
+            # 持久化任务记录（便于回溯/重试）
+            _persist_job(
+                task.id,
+                request.project_id,
+                status="queued",
+                progress=0.0,
+                message="合成任务已提交到队列",
+                request=request.model_dump(),
+            )
+            return ComposeResponse(
+                success=True,
+                message="合成任务已提交到队列",
+                task_id=task.id
+            )
+        except Exception as e:
+            logger.warning(f"Celery 任务提交失败，降级为同步模式: {e}")
+    
+    # 降级：使用 FastAPI BackgroundTasks
+    task_id = f"compose_{request.project_id}_{int(time.time())}"
+    _persist_job(task_id, request.project_id, status="queued", progress=0.0, message="任务已创建（本地模式）", request=request.model_dump())
+    
+    background_tasks.add_task(
+        _do_compose,
+        task_id,
+        request
+    )
+    
+    return ComposeResponse(
+        success=True,
+        message="合成任务已提交",
+        task_id=task_id
+    )
+
+
+@router.post("/render-sync", response_model=ComposeResponse)
+async def render_video_sync(request: ComposeRequest):
+    """
+    同步合成视频（适合短视频）
+    直接返回结果
+    """
+    project = db.get_project(request.project_id)
+    if not project:
+        raise HTTPException(status_code=404, detail="项目不存在")
+    
+    try:
+        # 轻量导出前校验（MVP）：禁止同轨道视频重叠
+        if request.video_clips:
+            _validate_no_overlap(request.video_clips)
+            _validate_trim_bounds(request.video_clips)
+        output_path = await _compose_video(None, request)
+        
+        return ComposeResponse(
+            success=True,
+            message="合成完成",
+            output_path=output_path,
+            output_url=f"/static/output/{Path(output_path).name}"
+        )
+        
+    except Exception as e:
+        logger.error(f"合成失败: {e}")
+        raise HTTPException(status_code=500, detail=str(e))
+
+
+@router.get("/status/{task_id}")
+async def get_compose_status(task_id: str):
+    """
+    获取合成任务状态
+    支持 Celery 任务和本地 BackgroundTask 两种模式
+    """
+    # 先检查是否是 Celery 任务
+    if CELERY_AVAILABLE:
+        try:
+            from celery.result import AsyncResult
+            from api.celery_app import celery_app
+            
+            result = AsyncResult(task_id, app=celery_app)
+            
+            if result.state == "PENDING":
+                _persist_job(task_id, (db.get_render_job(task_id) or {}).get("project_id") or "", status="pending", progress=0.0, message="等待处理...")
+                return ComposeStatus(status="pending", progress=0, message="等待处理...")
+            elif result.state == "PROGRESS":
+                meta = result.info or {}
+                _persist_job(task_id, (db.get_render_job(task_id) or {}).get("project_id") or "", status="processing", progress=meta.get("progress", 0), message=meta.get("message", "处理中..."))
+                return ComposeStatus(
+                    status="processing",
+                    progress=meta.get("progress", 0),
+                    message=meta.get("message", "处理中...")
+                )
+            elif result.state == "SUCCESS":
+                data = result.result or {}
+                _persist_job(task_id, (db.get_render_job(task_id) or {}).get("project_id") or "", status="completed", progress=1.0, message="合成完成")
+                try:
+                    db.update_render_job(task_id, {"output_path": data.get("output_path"), "output_url": data.get("output_url")})
+                except Exception:
+                    pass
+                return ComposeStatus(
+                    status="completed",
+                    progress=1.0,
+                    message="合成完成",
+                    output_path=data.get("output_path"),
+                    output_url=data.get("output_url")
+                )
+            elif result.state == "FAILURE":
+                _persist_job(task_id, (db.get_render_job(task_id) or {}).get("project_id") or "", status="failed", progress=0.0, message=str(result.info) if result.info else "任务失败")
+                try:
+                    db.update_render_job(task_id, {"error": str(result.info) if result.info else "任务失败"})
+                except Exception:
+                    pass
+                return ComposeStatus(
+                    status="failed",
+                    progress=0,
+                    message=str(result.info) if result.info else "任务失败"
+                )
+            else:
+                return ComposeStatus(status=result.state.lower(), progress=0, message=result.state)
+                
+        except Exception as e:
+            logger.debug(f"Celery 状态查询失败: {e}")
+    
+    # 本地任务/持久化任务：从 DB 取
+    job = db.get_render_job(task_id)
+    if job:
+        return ComposeStatus(
+            status=job.get("status") or "pending",
+            progress=float(job.get("progress") or 0.0),
+            message=job.get("message") or "",
+            output_path=job.get("output_path"),
+            output_url=job.get("output_url"),
+        )
+    
+    raise HTTPException(status_code=404, detail="任务不存在")
+
+
+@router.post("/retry/{task_id}", response_model=ComposeResponse)
+async def retry_compose(task_id: str, background_tasks: BackgroundTasks):
+    """失败任务一键重试：复用原 request 创建新任务。"""
+    job = db.get_render_job(task_id)
+    if not job or not job.get("request"):
+        raise HTTPException(status_code=404, detail="任务不存在或无可重试的请求数据")
+    if (job.get("status") or "").lower() not in ("failed", "cancelled"):
+        raise HTTPException(status_code=400, detail="仅失败/取消的任务可重试")
+
+    req = job["request"]
+    new_id = f"compose_{job.get('project_id')}_{int(time.time())}"
+    db.create_render_job(new_id, job.get("project_id") or "", status="queued", progress=0.0, message="重试任务已创建", request=req, parent_id=task_id)
+    # 走 BackgroundTasks（不阻塞请求）
+    background_tasks.add_task(_do_compose, new_id, ComposeRequest(**req))
+    return ComposeResponse(success=True, message="重试任务已提交", task_id=new_id)
+
+
+@router.post("/quick", response_model=ComposeResponse)
+async def quick_compose(project_id: str, bgm_id: Optional[str] = None, async_mode: bool = False):
+    """
+    快速合成（使用项目默认设置）
+    适合工作流一键合成
+    
+    Args:
+        project_id: 项目 ID
+        bgm_id: BGM 文件名
+        async_mode: 是否使用异步模式（推荐大量并发时使用）
+    """
+    project = db.get_project(project_id)
+    if not project:
+        raise HTTPException(status_code=404, detail="项目不存在")
+    
+    script_data = project.get("script_data")
+    if not script_data:
+        raise HTTPException(status_code=400, detail="项目缺少脚本数据")
+    
+    # 获取视频素材
+    assets = db.get_assets(project_id, "video")
+    video_map = {a["scene_id"]: a["local_path"] for a in assets if a["status"] == "completed"}
+    
+    if not video_map:
+        raise HTTPException(status_code=400, detail="项目缺少视频素材")
+    
+    # BGM 路径
+    bgm_path = None
+    if bgm_id:
+        bgm_path = str(config.ASSETS_DIR / "bgm" / bgm_id)
+        if not os.path.exists(bgm_path):
+            bgm_path = None
+    
+    # 异步模式：提交到 Celery 队列
+    if async_mode and CELERY_AVAILABLE:
+        try:
+            task = compose_from_script_task.delay(
+                project_id=project_id,
+                script_data=script_data,
+                video_map=video_map,
+                bgm_path=bgm_path,
+                voice_type=config.VOLC_TTS_DEFAULT_VOICE
+            )
+            
+            return ComposeResponse(
+                success=True,
+                message="合成任务已提交到队列",
+                task_id=task.id
+            )
+        except Exception as e:
+            logger.warning(f"Celery 任务提交失败，降级为同步模式: {e}")
+    
+    # 同步模式
+    try:
+        composer = VideoComposer(voice_type=config.VOLC_TTS_DEFAULT_VOICE)
+        
+        output_name = f"final_{project_id}_{int(time.time())}"
+        output_path = composer.compose_from_script(
+            script=script_data,
+            video_map=video_map,
+            bgm_path=bgm_path,
+            output_name=output_name
+        )
+        
+        # 保存到数据库
+        db.save_asset(project_id, 0, "final_video", "completed", local_path=output_path)
+        db.update_project_status(project_id, "completed")
+        
+        return ComposeResponse(
+            success=True,
+            message="合成完成",
+            output_path=output_path,
+            output_url=f"/static/output/{Path(output_path).name}"
+        )
+        
+    except Exception as e:
+        logger.error(f"快速合成失败: {e}")
+        raise HTTPException(status_code=500, detail=str(e))
+
+
+# ============================================================
+# Internal Functions
+# ============================================================
+
+async def _do_compose(task_id: str, request: ComposeRequest):
+    """后台合成任务"""
+    try:
+        db.update_render_job(task_id, {"status": "processing", "progress": 0.05, "message": "正在准备素材..."})
+        
+        output_path = await _compose_video(task_id, request)
+        
+        db.update_render_job(task_id, {
+            "status": "completed",
+            "progress": 1.0,
+            "message": "合成完成",
+            "output_path": output_path,
+            "output_url": f"/static/output/{Path(output_path).name}",
+        })
+        
+        # 保存到数据库
+        db.save_asset(request.project_id, 0, "final_video", "completed", local_path=output_path)
+        db.update_project_status(request.project_id, "completed")
+        
+    except Exception as e:
+        logger.error(f"合成任务失败: {e}")
+        db.update_render_job(task_id, {"status": "failed", "progress": 0.0, "message": str(e), "error": str(e)})
+
+
+async def _compose_video(task_id: Optional[str], request: ComposeRequest) -> str:
+    """
+    执行视频合成
+    基于编辑器轨道数据
+    """
+    timestamp = int(time.time())
+    output_name = request.output_name or f"composed_{request.project_id}_{timestamp}"
+    
+    if task_id:
+        db.update_render_job(task_id, {"status": "processing", "progress": 0.10, "message": "收集视频片段..."})
+    # 1. 收集视频片段（按时间排序）
+    video_clips = sorted(request.video_clips, key=lambda x: x.get("start", 0))
+    
+    if not video_clips:
+        raise ValueError("没有视频片段")
+    
+    video_paths = []
+    video_fades = []
+    for clip in video_clips:
+        source_path = clip.get("source_path")
+        if source_path and os.path.exists(source_path):
+            # 如果有裁剪，先裁剪
+            trim_start = clip.get("trim_start", 0)
+            trim_end = clip.get("trim_end")
+            
+            # 读取转场参数（存放在 style 里；必须 WYSIWYG：预览=导出）
+            style = clip.get("style") or {}
+            vf_in = float(style.get("vFadeIn") or style.get("v_fade_in") or 0.0)
+            vf_out = float(style.get("vFadeOut") or style.get("v_fade_out") or 0.0)
+            # 转场库（WYSIWYG）：仅使用 vTransitionType/vTransitionDur，由 ffmpeg_utils 在拼接阶段实现
+            t_type = str(style.get("vTransitionType") or style.get("v_transition_type") or "")
+            try:
+                t_dur = float(style.get("vTransitionDur") or style.get("v_transition_dur") or 0.0)
+            except Exception:
+                t_dur = 0.0
+
+            if trim_start > 0 or trim_end:
+                # 需要裁剪
+                trimmed_path = str(config.TEMP_DIR / f"trim_{timestamp}_{len(video_paths)}.mp4")
+                duration = (trim_end or 999) - trim_start
+                cmd = [
+                    ffmpeg_utils.FFMPEG_PATH, "-y",
+                    "-ss", str(trim_start),
+                    "-i", source_path,
+                    "-t", str(duration),
+                    "-c", "copy",
+                    trimmed_path
+                ]
+                ffmpeg_utils._run_ffmpeg(cmd)
+                video_paths.append(trimmed_path)
+                video_fades.append({"in": max(0.0, vf_in), "out": max(0.0, vf_out), "type": t_type, "dur": max(0.0, t_dur)})
+            else:
+                video_paths.append(source_path)
+                video_fades.append({"in": max(0.0, vf_in), "out": max(0.0, vf_out), "type": t_type, "dur": max(0.0, t_dur)})
+    
+    # 2. 拼接视频（尽量保留原声；失败则回退无音频）
+    if task_id:
+        db.update_render_job(task_id, {"progress": 0.25, "message": "拼接视频..."})
+    merged_path = str(config.TEMP_DIR / f"{output_name}_merged.mp4")
+    ffmpeg_utils.concat_videos_with_audio(video_paths, merged_path, (1080, 1920), fades=video_fades)
+    current_video = merged_path
+    
+    # 添加静音轨（仅当拼接结果没有音轨时）
+    if task_id:
+        db.update_render_job(task_id, {"progress": 0.30, "message": "添加静音轨..."})
+    has_audio = False
+    try:
+        r = subprocess.run(
+            [ffmpeg_utils.FFPROBE_PATH, "-v", "error", "-select_streams", "a", "-show_entries", "stream=index", "-of", "csv=p=0", current_video],
+            capture_output=True,
+            text=True,
+        )
+        has_audio = bool((r.stdout or "").strip())
+    except Exception:
+        has_audio = False
+    if not has_audio:
+        silent_path = str(config.TEMP_DIR / f"{output_name}_silent.mp4")
+        ffmpeg_utils.add_silence_audio(current_video, silent_path)
+        current_video = silent_path
+    
+    # 获取总时长
+    info = ffmpeg_utils.get_video_info(current_video)
+    total_duration = float(info.get("duration", 10))
+    
+    # 3. 生成并混入旁白
+    if request.voiceover_clips:
+        if task_id:
+            db.update_render_job(task_id, {"progress": 0.38, "message": "生成/混入旁白..."})
+        mixed_audio_path = str(config.TEMP_DIR / f"{output_name}_mixed_vo.mp3")
+        
+        # 初始化静音底轨
+        ffmpeg_utils._run_ffmpeg([
+            ffmpeg_utils.FFMPEG_PATH, "-y",
+            "-f", "lavfi", "-i", "anullsrc=r=44100:cl=stereo",
+            "-t", str(total_duration),
+            "-c:a", "mp3",
+            mixed_audio_path
+        ])
+        
+        for i, clip in enumerate(request.voiceover_clips):
+            start_time = float(clip.get("start", 0) or 0.0)
+            target_duration = float(clip.get("duration", 3) or 3.0)
+            vol = float(clip.get("volume", 1.0) or 1.0)
+            fade_in = float(clip.get("fade_in", 0.05) or 0.0)
+            fade_out = float(clip.get("fade_out", 0.05) or 0.0)
+            playback_rate = clip.get("playback_rate", clip.get("playbackRate", None))
+            try:
+                playback_rate = float(playback_rate) if playback_rate is not None else None
+            except Exception:
+                playback_rate = None
+
+            # 优先使用已生成的旁白音频（编辑器预览生成）
+            existing_path = clip.get("source_path")
+            voice_path = None
+            if existing_path and os.path.exists(existing_path):
+                voice_path = existing_path
+            else:
+                text = (clip.get("text") or "").strip()
+                if not text:
+                    continue
+                voice_path = factory.generate_voiceover_volcengine(
+                    text=text,
+                    voice_type=request.voice_type,
+                    output_path=str(config.TEMP_DIR / f"{output_name}_vo_{i}.mp3")
+                )
+                if not voice_path:
+                    continue
+
+            # 旁白：纯播放倍速贴合时长（可快可慢，预览/导出一致）
+            adjusted_path = str(config.TEMP_DIR / f"{output_name}_vo_adj_{i}.mp3")
+            if playback_rate and playback_rate > 0:
+                # 先按用户倍速改变语速，再严格裁剪/补齐到片段时长（不再二次计算倍速）
+                sped = str(config.TEMP_DIR / f"{output_name}_vo_sp_{i}.mp3")
+                ffmpeg_utils.change_audio_speed(voice_path, playback_rate, sped)
+                ffmpeg_utils.force_audio_duration(sped, target_duration, adjusted_path)
+            else:
+                ffmpeg_utils.fit_audio_to_duration_by_speed(voice_path, target_duration, adjusted_path)
+
+            # 音量/淡入淡出（与预览一致）
+            processed_path = adjusted_path
+            try:
+                if vol != 1.0 or fade_in > 0 or fade_out > 0:
+                    processed_path = str(config.TEMP_DIR / f"{output_name}_vo_fx_{i}.mp3")
+                    fo_st = max(target_duration - fade_out, 0.0)
+                    af = (
+                        f"afade=t=in:st=0:d={fade_in},"
+                        f"afade=t=out:st={fo_st}:d={fade_out},"
+                        f"volume={vol}"
+                    )
+                    ffmpeg_utils._run_ffmpeg([
+                        ffmpeg_utils.FFMPEG_PATH, "-y",
+                        "-i", adjusted_path,
+                        "-filter:a", af,
+                        processed_path
+                    ])
+            except Exception:
+                processed_path = adjusted_path
+
+            # 混合
+            new_mixed = str(config.TEMP_DIR / f"{output_name}_mixed_{i}.mp3")
+            ffmpeg_utils.mix_audio_at_offset(mixed_audio_path, processed_path, start_time, new_mixed, overlay_volume=1.0)
+            mixed_audio_path = new_mixed
+        
+        # 混入视频
+        if task_id:
+            db.update_render_job(task_id, {"progress": 0.55, "message": "混音合成..."})
+        voiced_path = str(config.TEMP_DIR / f"{output_name}_voiced.mp4")
+        ffmpeg_utils.mix_audio(
+            current_video, mixed_audio_path, voiced_path,
+            audio_volume=1.5,
+            video_volume=0.2
+        )
+        current_video = voiced_path
+    
+    # 4. 添加字幕（使用文本渲染器生成 PNG 叠加，支持字体/颜色/B/I/U）
+    if request.subtitle_clips:
+        if task_id:
+            db.update_render_job(task_id, {"progress": 0.65, "message": "渲染字幕..."})
+        overlay_configs = []
+        for clip in request.subtitle_clips:
+            text = (clip.get("text") or "").strip()
+            if not text:
+                continue
+            style = clip.get("style") or {}
+            # WYSIWYG：自动换行交给 TextRenderer（按 max_width 控制）
+
+            # 将前端 style 映射到 renderer style
+            r_style = {
+                "font_family": style.get("font_family") or style.get("fontFamily") or None,
+                "font_size": int(style.get("font_size") or style.get("fontSize") or 60),
+                "font_color": style.get("font_color") or style.get("fontColor") or "#FFFFFF",
+                "bold": bool(style.get("bold", False)),
+                "italic": bool(style.get("italic", False)),
+                "underline": bool(style.get("underline", False)),
+                # 默认给字幕一点描边，提升可读性
+                "stroke": style.get("stroke") or {"color": "#000000", "width": int(style.get("stroke_width") or 5)},
+            }
+
+            # 文本框宽度：前端用 0~1（相对画布宽度）控制换行；导出映射到像素（1080）
+            try:
+                box_w = style.get("box_w") or style.get("boxW") or style.get("max_width") or style.get("maxWidth")
+                if isinstance(box_w, (int, float)):
+                    if 0 < float(box_w) <= 1:
+                        r_style["max_width"] = int(1080 * float(box_w))
+                    elif float(box_w) > 1:
+                        r_style["max_width"] = int(float(box_w))
+            except Exception:
+                pass
+
+            img_path = renderer.render(text, r_style, cache=False)
+
+            def _pos(v, axis: str):
+                # WYSIWYG：前端 position 允许 0~1（相对画面）或 ffmpeg 表达式字符串
+                if isinstance(v, (int, float)):
+                    fv = float(v)
+                    if 0 <= fv <= 1:
+                        return f"({axis}-w)*{fv}" if axis == "W" else f"({axis}-h)*{fv}"
+                    return str(fv)
+                if isinstance(v, str) and v.strip():
+                    return v
+                return None
+
+            position = clip.get("position") or {}
+            overlay_configs.append({
+                "path": img_path,
+                "x": _pos(position.get("x"), "W") or "(W-w)/2",
+                "y": _pos(position.get("y"), "H") or "H*0.78",
+                "start": clip.get("start", 0),
+                "duration": clip.get("duration", 3)
+            })
+
+        if overlay_configs:
+            subtitled_path = str(config.TEMP_DIR / f"{output_name}_subtitled.mp4")
+            ffmpeg_utils.overlay_multiple_images(current_video, overlay_configs, subtitled_path)
+            current_video = subtitled_path
+    
+    # 5. 叠加花字
+    if request.fancy_text_clips:
+        if task_id:
+            db.update_render_job(task_id, {"progress": 0.78, "message": "叠加花字..."})
+        overlay_configs = []
+        for clip in request.fancy_text_clips:
+            text = clip.get("text", "")
+            if not text:
+                continue
+            
+            style = clip.get("style", {
+                "font_size": 72,
+                "font_color": "#FFFFFF",
+                "stroke": {"color": "#000000", "width": 5}
+            })
+
+            # 文本框宽度：0~1（相对画布宽度）-> 像素（1080）
+            try:
+                box_w = style.get("box_w") or style.get("boxW") or style.get("max_width") or style.get("maxWidth")
+                if isinstance(box_w, (int, float)):
+                    if 0 < float(box_w) <= 1:
+                        style["max_width"] = int(1080 * float(box_w))
+                    elif float(box_w) > 1:
+                        style["max_width"] = int(float(box_w))
+            except Exception:
+                pass
+            
+            img_path = renderer.render(text, style, cache=False)
+            
+            def _pos(v, axis: str):
+                if isinstance(v, (int, float)):
+                    fv = float(v)
+                    if 0 <= fv <= 1:
+                        return f"({axis}-w)*{fv}" if axis == "W" else f"({axis}-h)*{fv}"
+                    return str(fv)
+                if isinstance(v, str) and v.strip():
+                    return v
+                return None
+
+            position = clip.get("position", {})
+            overlay_configs.append({
+                "path": img_path,
+                "x": _pos(position.get("x"), "W") or "(W-w)/2",
+                "y": _pos(position.get("y"), "H") or "180",
+                "start": clip.get("start", 0),
+                "duration": clip.get("duration", 5)
+            })
+        
+        if overlay_configs:
+            fancy_path = str(config.TEMP_DIR / f"{output_name}_fancy.mp4")
+            ffmpeg_utils.overlay_multiple_images(current_video, overlay_configs, fancy_path)
+            current_video = fancy_path
+
+    # 5.5 叠加贴纸（PNG/SVG）
+    if request.sticker_clips:
+        if task_id:
+            db.update_render_job(task_id, {"progress": 0.82, "message": "叠加贴纸..."})
+        overlay_configs = []
+        for i, clip in enumerate(request.sticker_clips):
+            src_url = clip.get("source_url") or clip.get("sourceUrl")
+            src_path = clip.get("source_path") or clip.get("sourcePath")
+            # 贴纸可能来自 /static/assets/...，导出侧优先用 source_path，其次尝试解析 url 到本地路径
+            path = src_path
+            if not path and isinstance(src_url, str) and src_url.startswith("/static/"):
+                # /static/assets/... 映射到容器内 /app/assets/...
+                try:
+                    rel = src_url.replace("/static/assets/", "")
+                    path = str(config.ASSETS_DIR / rel)
+                except Exception:
+                    path = None
+            if not path or not os.path.exists(path):
+                continue
+
+            # 规范化为 PNG
+            png_path = str(config.TEMP_DIR / f"{output_name}_st_{i}.png")
+            try:
+                norm = ffmpeg_utils.normalize_sticker_to_png(path, png_path)
+            except Exception:
+                norm = None
+            if not norm or not os.path.exists(norm):
+                continue
+
+            pos = clip.get("position") or {}
+            x = pos.get("x", 0.8)
+            y = pos.get("y", 0.2)
+            # x/y: 0~1 视为比例
+            def _pos(v, axis: str):
+                if isinstance(v, (int, float)):
+                    fv = float(v)
+                    if 0 <= fv <= 1:
+                        return f"({axis}-w)*{fv}" if axis == "W" else f"({axis}-h)*{fv}"
+                    return str(fv)
+                if isinstance(v, str) and v.strip():
+                    return v
+                return None
+
+            st = clip.get("style") or {}
+            scale = 1.0
+            try:
+                scale = float(st.get("scale", 1.0) or 1.0)
+            except Exception:
+                scale = 1.0
+            scale = max(0.3, min(3.0, scale))
+
+            # 通过 overlay 的 scale：先 scale 贴纸，再 overlay
+            # overlay_multiple_images 不支持单张单独 scale filter，所以这里把 PNG 预缩放成新文件
+            scaled_path = str(config.TEMP_DIR / f"{output_name}_st_sc_{i}.png")
+            try:
+                if abs(scale - 1.0) > 1e-3:
+                    ffmpeg_utils._run_ffmpeg([
+                        ffmpeg_utils.FFMPEG_PATH, "-y",
+                        "-i", norm,
+                        "-vf", f"scale=iw*{scale}:ih*{scale}:flags=lanczos",
+                        scaled_path
+                    ])
+                    norm = scaled_path
+            except Exception:
+                pass
+
+            overlay_configs.append({
+                "path": norm,
+                "x": _pos(x, "W") or "(W-w)*0.8",
+                "y": _pos(y, "H") or "(H-h)*0.2",
+                "start": clip.get("start", 0),
+                "duration": clip.get("duration", 2),
+            })
+
+        if overlay_configs:
+            st_path = str(config.TEMP_DIR / f"{output_name}_stickers.mp4")
+            ffmpeg_utils.overlay_multiple_images(current_video, overlay_configs, st_path)
+            current_video = st_path
+    
+    # 6. 添加 BGM
+    if request.bgm_clip:
+        if task_id:
+            db.update_render_job(task_id, {"progress": 0.88, "message": "混入 BGM..."})
+        bgm_source = request.bgm_clip.get("source_path")
+        if bgm_source and os.path.exists(bgm_source):
+            bgm_output = str(config.TEMP_DIR / f"{output_name}_bgm.mp4")
+            # BGM 参数：优先使用 clip 内配置
+            bgm_start = float(request.bgm_clip.get("start", 0) or 0.0)
+            bgm_dur = request.bgm_clip.get("duration")
+            try:
+                bgm_dur = float(bgm_dur) if bgm_dur is not None else None
+            except Exception:
+                bgm_dur = None
+            bgm_vol = float(request.bgm_clip.get("volume", request.bgm_volume) or request.bgm_volume)
+            bgm_fade_in = float(request.bgm_clip.get("fade_in", 0.8) or 0.0)
+            bgm_fade_out = float(request.bgm_clip.get("fade_out", 0.8) or 0.0)
+            bgm_ducking = request.bgm_clip.get("ducking")
+            if not isinstance(bgm_ducking, bool):
+                bgm_ducking = True
+            bgm_duck_volume = float(request.bgm_clip.get("duck_volume", 0.25) or 0.25)
+
+            # 闪避区间：来自旁白时间轴（更可控）
+            duck_ranges = None
+            if request.voiceover_clips and bgm_ducking:
+                duck_ranges = []
+                for vc in request.voiceover_clips:
+                    s = float(vc.get("start", 0) or 0.0)
+                    d = float(vc.get("duration", 0) or 0.0)
+                    if d <= 0:
+                        continue
+                    # 给一点点缓冲，避免边缘咬字突兀
+                    duck_ranges.append((max(0.0, s - 0.03), s + d + 0.05))
+
+            ffmpeg_utils.add_bgm(
+                current_video, bgm_source, bgm_output,
+                bgm_volume=bgm_vol,
+                ducking=bool(bgm_ducking),
+                duck_volume=bgm_duck_volume,
+                duck_ranges=duck_ranges,
+                start_time=bgm_start,
+                clip_duration=bgm_dur,
+                fade_in=bgm_fade_in,
+                fade_out=bgm_fade_out,
+            )
+            current_video = bgm_output
+    
+    # 7. 输出最终文件
+    if task_id:
+        db.update_render_job(task_id, {"progress": 0.95, "message": "写出最终文件..."})
+    final_path = str(config.OUTPUT_DIR / f"{output_name}.mp4")
+    import shutil
+    shutil.copy(current_video, final_path)
+    
+    logger.info(f"合成完成: {final_path}")
+    return final_path
+

api/routes/editor.py

@@ -0,0 +1,825 @@
+"""
+编辑器 API 路由
+提供时间轴编辑、轨道管理功能
+"""
+import os
+import time
+import json
+import logging
+from typing import List, Optional, Dict, Any
+from pathlib import Path
+from functools import lru_cache
+
+from fastapi import APIRouter, HTTPException, Body
+from pydantic import BaseModel, Field
+
+import config
+from modules.db_manager import db
+from modules import factory, ffmpeg_utils
+from modules.text_renderer import renderer
+from modules.legacy_path_mapper import map_legacy_local_path
+from modules.preview_proxy import ensure_video_proxy
+
+logger = logging.getLogger(__name__)
+router = APIRouter()
+
+def _safe_float(v: Any) -> Optional[float]:
+    try:
+        if v is None:
+            return None
+        return float(v)
+    except Exception:
+        return None
+
+
+@lru_cache(maxsize=2048)
+def _probe_source_duration_cache_key(path: str, mtime_ns: int, size: int) -> float:
+    """
+    Cache wrapper to avoid repeated ffprobe for same file version.
+    Note: caller must pass (path, stat.mtime_ns, stat.size).
+    """
+    info = ffmpeg_utils.get_video_info(path)
+    return float(info.get("duration") or 0.0)
+
+
+def _get_source_duration_seconds(path: Optional[str]) -> Optional[float]:
+    if not path:
+        return None
+    if not os.path.exists(path):
+        return None
+    try:
+        st = os.stat(path)
+        dur = _probe_source_duration_cache_key(path, st.st_mtime_ns, st.st_size)
+        return dur if dur and dur > 0 else None
+    except Exception:
+        return None
+
+
+# ============================================================
+# Pydantic Models - 编辑器数据结构
+# ============================================================
+
+class TimelineClip(BaseModel):
+    """时间轴片段"""
+    id: str
+    type: str  # video, audio, subtitle, fancy_text, bgm
+    start: float  # 开始时间（秒）
+    duration: float  # 持续时间（秒）
+    source_path: Optional[str] = None  # 源文件路径
+    source_url: Optional[str] = None  # 源文件 URL
+    
+    # 视频特有
+    trim_start: float = 0  # 裁剪起点
+    trim_end: Optional[float] = None  # 裁剪终点
+    source_duration: Optional[float] = None  # 源素材时长（秒）
+    
+    # 文本特有
+    text: Optional[str] = None
+    style: Optional[Dict[str, Any]] = None
+    position: Optional[Dict[str, Any]] = None  # {x, y}
+    
+    # 音频特有
+    volume: float = 1.0
+    fade_in: Optional[float] = None
+    fade_out: Optional[float] = None
+    ducking: Optional[bool] = None
+    duck_volume: Optional[float] = None
+    playback_rate: Optional[float] = None
+
+
+class Track(BaseModel):
+    """轨道"""
+    id: str
+    name: str
+    type: str  # video, audio, subtitle, fancy_text, bgm, sticker
+    clips: List[TimelineClip] = []
+    locked: bool = False
+    visible: bool = True
+    muted: bool = False
+
+
+class EditorState(BaseModel):
+    """编辑器状态"""
+    project_id: str
+    total_duration: float = 0
+    tracks: List[Track] = []
+    current_time: float = 0
+    zoom: float = 1.0
+    ripple_mode: bool = True
+    subtitle_style: Optional[Dict[str, Any]] = None
+
+
+class VoiceoverRequest(BaseModel):
+    """旁白生成请求"""
+    text: str
+    voice_type: str = "zh_female_santongyongns_saturn_bigtts"
+    target_duration: Optional[float] = None
+
+
+class FancyTextRequest(BaseModel):
+    """花字生成请求"""
+    text: str
+    style: Dict[str, Any] = Field(default_factory=lambda: {
+        "font_size": 72,
+        "font_color": "#FFFFFF",
+        "stroke": {"color": "#000000", "width": 5}
+    })
+
+
+class TrimRequest(BaseModel):
+    """视频裁剪请求"""
+    source_path: str
+    start_time: float
+    end_time: float
+
+
+# ============================================================
+# API Endpoints
+# ============================================================
+
+@router.get("/{project_id}/state", response_model=EditorState)
+async def get_editor_state(project_id: str, use_proxy: bool = True):
+    """
+    获取编辑器状态
+    从项目数据和素材自动构建时间轴
+    """
+    project = db.get_project(project_id)
+    if not project:
+        raise HTTPException(status_code=404, detail="项目不存在")
+    
+    # 获取素材
+    assets = db.get_assets(project_id)
+    script_data = project.get("script_data", {})
+    product_info = project.get("product_info", {}) or {}
+
+    # 如果已保存过 editor_state，优先回放（非破坏式：轨道时间轴优先）
+    saved_state = product_info.get("editor_state")
+    if isinstance(saved_state, dict) and saved_state.get("tracks"):
+        try:
+            state_obj = EditorState.model_validate(saved_state)  # pydantic v2
+
+            # 回放兜底补全：source_duration 可能是旧版本保存为 null
+            for t in state_obj.tracks:
+                for c in t.clips:
+                    # 统一用 source_duration 表示“源素材时长”（视频/音频都适用）
+                    if c.source_duration is None or (isinstance(c.source_duration, (int, float)) and c.source_duration <= 0):
+                        if c.source_path:
+                            try:
+                                if t.type in ("audio", "bgm") or c.type in ("audio", "bgm"):
+                                    c.source_duration = float(ffmpeg_utils.get_audio_info(c.source_path).get("duration") or 0) or None
+                                else:
+                                    c.source_duration = _get_source_duration_seconds(c.source_path)
+                            except Exception:
+                                c.source_duration = _get_source_duration_seconds(c.source_path)
+                    # normalize trim_end if missing（仅视频有意义，但这里兜底不会出错）
+                    if c.trim_end is None:
+                        c.trim_end = (c.trim_start or 0) + (c.duration or 0)
+
+            # 回放兜底补全：如果保存态里“旁白/字幕/花字”为空，但 script_data 里有内容，则自动回填
+            voiceover_timeline = (script_data or {}).get("voiceover_timeline") or []
+            scenes = (script_data or {}).get("scenes") or []
+
+            def _ensure_track(tid: str, ttype: str, name: str) -> Track:
+                existing = next((t for t in state_obj.tracks if t.id == tid), None)
+                if existing:
+                    return existing
+                t = Track(id=tid, name=name, type=ttype, clips=[])
+                state_obj.tracks.append(t)
+                return t
+
+            # ensure total_duration for ratio-derived items
+            total_duration = float(state_obj.total_duration or 0.0)
+
+            # voiceover
+            vo_t = _ensure_track("audio-voiceover", "audio", "旁白")
+            if (not vo_t.clips) and voiceover_timeline:
+                for i, item in enumerate(voiceover_timeline):
+                    start_time = float(item.get("start_time", item.get("start_ratio", 0) * total_duration))
+                    duration = float(item.get("duration", item.get("duration_ratio", 0.25) * total_duration))
+                    vo_t.clips.append(TimelineClip(
+                        id=f"vo-{i}",
+                        type="audio",
+                        start=start_time,
+                        duration=duration,
+                        text=item.get("text", ""),
+                        volume=1.0
+                    ))
+
+            # subtitle
+            sub_t = _ensure_track("subtitle-main", "subtitle", "字幕")
+            if (not sub_t.clips) and voiceover_timeline:
+                for i, item in enumerate(voiceover_timeline):
+                    start_time = float(item.get("start_time", item.get("start_ratio", 0) * total_duration))
+                    duration = float(item.get("duration", item.get("duration_ratio", 0.25) * total_duration))
+                    sub_t.clips.append(TimelineClip(
+                        id=f"sub-{i}",
+                        type="subtitle",
+                        start=start_time,
+                        duration=duration,
+                        text=item.get("subtitle", item.get("text", "")),
+                        style={"fontsize": 60, "fontcolor": "white"},
+                        position={"x": "(w-text_w)/2", "y": "h-200"}
+                    ))
+
+            # fancy text
+            fancy_t = _ensure_track("fancy-text", "fancy_text", "花字")
+            if (not fancy_t.clips) and scenes:
+                # best effort: align with existing video duration if any
+                scene_start = 0.0
+                video_track = next((t for t in state_obj.tracks if t.type == "video"), None)
+                # build scene durations from video clips if possible
+                scene_durations = []
+                if video_track and video_track.clips:
+                    scene_durations = [float(c.duration or 0.0) for c in video_track.clips]
+                for idx, scene in enumerate(scenes):
+                    scene_duration = scene_durations[idx] if idx < len(scene_durations) and scene_durations[idx] > 0 else 5.0
+                    ft = scene.get("fancy_text", {}) if isinstance(scene, dict) else {}
+                    if isinstance(ft, dict) and ft.get("text"):
+                        fancy_t.clips.append(TimelineClip(
+                            id=f"fancy-{scene.get('id', idx)}",
+                            type="fancy_text",
+                            start=scene_start,
+                            duration=scene_duration,
+                            text=ft.get("text", ""),
+                            style={
+                                "font_size": 72,
+                                "font_color": "#FFFFFF",
+                                "stroke": {"color": "#000000", "width": 5}
+                            },
+                            position={"x": "(W-w)/2", "y": "180"}
+                        ))
+                    scene_start += scene_duration
+            # stickers（贴纸轨道：默认存在，便于拖拽添加）
+            _ensure_track("sticker-main", "sticker", "贴纸")
+
+            # video（如果保存态里没有视频轨或为空，但 assets 里有视频，则回填，避免“从 video flow 进来却黑屏”）
+            video_assets = sorted(
+                [a for a in assets if a.get("asset_type") == "video" and a.get("status") == "completed"],
+                key=lambda x: x.get("scene_id", 0)
+            )
+            video_t = _ensure_track("video-main", "video", "视频")
+            if (not video_t.clips) and video_assets:
+                cur_t = 0.0
+                for asset in video_assets:
+                    remote_url = asset.get("remote_url")
+                    source_path, _ = map_legacy_local_path(asset.get("local_path"))
+                    duration = 5.0
+                    if source_path and os.path.exists(source_path):
+                        try:
+                            duration = float(ffmpeg_utils.get_video_info(source_path).get("duration", 5.0))
+                        except Exception:
+                            duration = 5.0
+                        # 统一走 file proxy（内部会处理 legacy 映射 / remote_url）
+                        url = f"/api/assets/file/{asset.get('id')}"
+                        video_t.clips.append(TimelineClip(
+                            id=f"video-{asset.get('scene_id')}",
+                            type="video",
+                            start=cur_t,
+                            duration=duration,
+                            source_path=source_path,
+                            source_url=url,
+                            trim_start=0,
+                            trim_end=duration,
+                            source_duration=duration
+                        ))
+                        cur_t += duration
+                        continue
+                    if remote_url and isinstance(remote_url, str) and remote_url.strip():
+                        meta = asset.get("metadata") or {}
+                        duration = float(meta.get("duration") or meta.get("source_duration") or 5.0)
+                        url = f"/api/assets/file/{asset.get('id')}"
+                        video_t.clips.append(TimelineClip(
+                            id=f"video-{asset.get('scene_id')}",
+                            type="video",
+                            start=cur_t,
+                            duration=duration,
+                            source_path=None,
+                            source_url=url,
+                            trim_start=0,
+                            trim_end=duration,
+                            source_duration=duration
+                        ))
+                        cur_t += duration
+
+            # ------------------------------
+            # 时间轴对齐：如果视频片段是真实时长（例如 3s），而字幕/花字按 5s 切段，
+            # 会出现“花字超出视频、拖拽时看不对齐”的体验。
+            # 这里按视频轨道重排花字/字幕/旁白的 start/duration，并把 total_duration 收敛到 videoEnd。
+            # ------------------------------
+            def _clip_end(c: TimelineClip) -> float:
+                try:
+                    return float(c.start or 0.0) + float(c.duration or 0.0)
+                except Exception:
+                    return 0.0
+
+            scene_timeline = []
+            if video_t and video_t.clips:
+                for vc in sorted(video_t.clips, key=lambda c: float(c.start or 0.0)):
+                    # prefer parse scene_id from id=video-{scene_id}
+                    sid = None
+                    if isinstance(vc.id, str) and vc.id.startswith("video-"):
+                        try:
+                            sid = int(vc.id.replace("video-", ""))
+                        except Exception:
+                            sid = None
+                    scene_timeline.append({
+                        "scene_id": sid,
+                        "start": float(vc.start or 0.0),
+                        "duration": max(0.01, float(vc.duration or 0.0)),
+                    })
+
+            if scene_timeline:
+                # 1) 花字：按 fancy-{scene_id} 精确对齐
+                fancy_t = next((t for t in state_obj.tracks if t.type == "fancy_text"), None)
+                if fancy_t and fancy_t.clips:
+                    by_id = {c.id: c for c in fancy_t.clips if isinstance(c.id, str)}
+                    for seg in scene_timeline:
+                        sid = seg["scene_id"]
+                        if sid is None:
+                            continue
+                        cid = f"fancy-{sid}"
+                        c = by_id.get(cid)
+                        if not c:
+                            continue
+                        c.start = float(seg["start"])
+                        c.duration = float(seg["duration"])
+                        # 修复“表达式坐标导致拖拽不直观”：初始化为居中百分比坐标（后续拖拽会改成数值）
+                        if isinstance(c.position, dict):
+                            if not isinstance(c.position.get("x"), (int, float)):
+                                c.position["x"] = 0.5
+                            if not isinstance(c.position.get("y"), (int, float)):
+                                c.position["y"] = 0.2
+                # 2) 字幕/旁白：如果片段数与场景数一致，则按索引对齐
+                subtitle_t = next((t for t in state_obj.tracks if t.type == "subtitle"), None)
+                voice_t = next((t for t in state_obj.tracks if t.id == "audio-voiceover"), None)
+                for tr in [subtitle_t, voice_t]:
+                    if not tr or not tr.clips:
+                        continue
+                    if len(tr.clips) != len(scene_timeline):
+                        continue
+                    for i, seg in enumerate(scene_timeline):
+                        tr.clips[i].start = float(seg["start"])
+                        tr.clips[i].duration = float(seg["duration"])
+
+                # 3) total_duration：收敛到视频结束时间
+                video_end = max(0.0, max((seg["start"] + seg["duration"]) for seg in scene_timeline))
+                if video_end > 0:
+                    state_obj.total_duration = float(video_end)
+
+                # 4) 兜底裁剪：任何片段不允许超出 total_duration（避免视频结束后黑屏但字幕/花字继续）
+                td = float(state_obj.total_duration or 0.0)
+                if td > 0:
+                    for tr in state_obj.tracks:
+                        kept = []
+                        for c in (tr.clips or []):
+                            if float(c.start or 0.0) >= td:
+                                continue
+                            end = _clip_end(c)
+                            if end > td:
+                                c.duration = max(0.01, td - float(c.start or 0.0))
+                            kept.append(c)
+                        tr.clips = kept
+
+            # ------------------------------
+            # BGM：如果没有片段，但脚本给了 bgm_style，则默认塞一条（可在前端再调整/替换）
+            # ------------------------------
+            bgm_t = next((t for t in state_obj.tracks if t.type == "bgm" or t.id == "audio-bgm"), None)
+            if bgm_t is None:
+                bgm_t = _ensure_track("audio-bgm", "bgm", "背景音乐")
+            if bgm_t and (not bgm_t.clips):
+                bgm_style = (script_data or {}).get("bgm_style") or ""
+                bgm_dir = config.ASSETS_DIR / "bgm"
+                chosen = None
+                if bgm_dir.exists():
+                    files = [f for f in bgm_dir.iterdir() if f.is_file() and f.suffix.lower() in [".mp3", ".mp4", ".m4a", ".wav"]]
+                    files.sort(key=lambda p: p.name)
+                    if isinstance(bgm_style, str) and bgm_style.strip():
+                        # very small heuristic: pick file that shares any keyword
+                        kws = [k.strip() for k in bgm_style.replace("，", " ").replace(",", " ").split() if len(k.strip()) >= 2]
+                        for f in files:
+                            name = f.stem
+                            if any(k in name for k in kws):
+                                chosen = f
+                                break
+                    if chosen is None and files:
+                        chosen = files[0]
+                if chosen is not None:
+                    td = float(state_obj.total_duration or 0.0)
+                    if td <= 0:
+                        # fallback: use max end across all tracks
+                        td = max(0.0, max((_clip_end(c) for t in state_obj.tracks for c in (t.clips or [])), default=0.0))
+                    if td > 0:
+                        bgm_t.clips.append(TimelineClip(
+                            id="bgm-0",
+                            type="bgm",
+                            start=0.0,
+                            duration=float(td),
+                            source_path=str(chosen),
+                            source_url=f"/static/assets/bgm/{chosen.name}",
+                            volume=0.25,
+                            style={"loop": True},
+                        ))
+            return state_obj
+        except Exception:
+            # fall back to rebuild
+            pass
+    
+    # 构建轨道
+    tracks = []
+    
+    # 1. 视频轨道
+    video_track = Track(
+        id="video-main",
+        name="视频",
+        type="video",
+        clips=[]
+    )
+    
+    current_time = 0
+    video_assets = sorted(
+        [a for a in assets if a["asset_type"] == "video" and a["status"] == "completed"],
+        key=lambda x: x["scene_id"]
+    )
+    
+    for asset in video_assets:
+        remote_url = asset.get("remote_url")
+        source_path, mapped_url = map_legacy_local_path(asset.get("local_path"))
+
+        # 1) 本地存在：正常走本地（统一用 /api/assets/file 作为 source_url，更稳）
+        if source_path and os.path.exists(source_path):
+            try:
+                info = ffmpeg_utils.get_video_info(source_path)
+                duration = float(info.get("duration", 5.0))
+            except Exception:
+                duration = 5.0
+
+            url = f"/api/assets/file/{asset['id']}"
+
+            video_track.clips.append(TimelineClip(
+                id=f"video-{asset['scene_id']}",
+                type="video",
+                start=current_time,
+                duration=duration,
+                source_path=source_path,
+                source_url=url,
+                trim_start=0,
+                trim_end=duration,
+                source_duration=duration
+            ))
+            current_time += duration
+            continue
+
+        # 2) 本地缺失但有 remote_url：也要能预览（至少不黑屏）
+        if remote_url and isinstance(remote_url, str) and remote_url.strip():
+            meta = asset.get("metadata") or {}
+            duration = float(meta.get("duration") or meta.get("source_duration") or 5.0)
+            url = f"/api/assets/file/{asset['id']}"  # 统一走 file proxy（会 307 到 remote_url）
+            video_track.clips.append(TimelineClip(
+                id=f"video-{asset['scene_id']}",
+                type="video",
+                start=current_time,
+                duration=duration,
+                source_path=None,
+                source_url=url,
+                trim_start=0,
+                trim_end=duration,
+                source_duration=duration
+            ))
+            current_time += duration
+    
+    tracks.append(video_track)
+    total_duration = current_time
+    
+    # 2. 旁白/TTS 轨道
+    voiceover_track = Track(
+        id="audio-voiceover",
+        name="旁白",
+        type="audio",
+        clips=[]
+    )
+    
+    voiceover_timeline = script_data.get("voiceover_timeline", [])
+    for i, item in enumerate(voiceover_timeline):
+        start_time = float(item.get("start_time", item.get("start_ratio", 0) * total_duration))
+        duration = float(item.get("duration", item.get("duration_ratio", 0.25) * total_duration))
+        
+        voiceover_track.clips.append(TimelineClip(
+            id=f"vo-{i}",
+            type="audio",
+            start=start_time,
+            duration=duration,
+            text=item.get("text", ""),
+            volume=1.0
+        ))
+    
+    tracks.append(voiceover_track)
+    
+    # 3. 字幕轨道
+    subtitle_track = Track(
+        id="subtitle-main",
+        name="字幕",
+        type="subtitle",
+        clips=[]
+    )
+    
+    for i, item in enumerate(voiceover_timeline):
+        start_time = float(item.get("start_time", item.get("start_ratio", 0) * total_duration))
+        duration = float(item.get("duration", item.get("duration_ratio", 0.25) * total_duration))
+        
+        subtitle_track.clips.append(TimelineClip(
+            id=f"sub-{i}",
+            type="subtitle",
+            start=start_time,
+            duration=duration,
+            text=item.get("subtitle", item.get("text", "")),
+            style={"fontsize": 60, "fontcolor": "white"},
+            position={"x": "(w-text_w)/2", "y": "h-200"}
+        ))
+    
+    tracks.append(subtitle_track)
+    
+    # 4. 花字轨道
+    fancy_track = Track(
+        id="fancy-text",
+        name="花字",
+        type="fancy_text",
+        clips=[]
+    )
+    
+    scenes = script_data.get("scenes", [])
+    scene_start = 0
+    for scene in scenes:
+        # 计算该场景的时长
+        scene_video = next(
+            (a for a in video_assets if a["scene_id"] == scene["id"]),
+            None
+        )
+        source_path, _ = map_legacy_local_path(scene_video.get("local_path") if scene_video else None)
+        if source_path and os.path.exists(source_path):
+            try:
+                info = ffmpeg_utils.get_video_info(source_path)
+                scene_duration = float(info.get("duration", 5.0))
+            except:
+                scene_duration = 5.0
+        else:
+            scene_duration = 5.0
+        
+        ft = scene.get("fancy_text", {})
+        if isinstance(ft, dict) and ft.get("text"):
+            fancy_track.clips.append(TimelineClip(
+                id=f"fancy-{scene['id']}",
+                type="fancy_text",
+                start=scene_start,
+                duration=scene_duration,
+                text=ft.get("text", ""),
+                style={
+                    "font_size": 72,
+                    "font_color": "#FFFFFF",
+                    "stroke": {"color": "#000000", "width": 5}
+                },
+                position={"x": "(W-w)/2", "y": "180"}
+            ))
+        
+        scene_start += scene_duration
+    
+    tracks.append(fancy_track)
+    
+    # 5. BGM 轨道
+    bgm_track = Track(
+        id="audio-bgm",
+        name="背景音乐",
+        type="bgm",
+        clips=[],
+        muted=False
+    )
+    # 默认 BGM：如果脚本给了 bgm_style，则塞一条，便于一键出片（用户可在前端替换/删除）
+    try:
+        bgm_style = (script_data or {}).get("bgm_style") or ""
+        bgm_dir = config.ASSETS_DIR / "bgm"
+        chosen = None
+        if bgm_dir.exists():
+            files = [f for f in bgm_dir.iterdir() if f.is_file() and f.suffix.lower() in [".mp3", ".mp4", ".m4a", ".wav"]]
+            files.sort(key=lambda p: p.name)
+            if isinstance(bgm_style, str) and bgm_style.strip():
+                kws = [k.strip() for k in bgm_style.replace("，", " ").replace(",", " ").split() if len(k.strip()) >= 2]
+                for f in files:
+                    name = f.stem
+                    if any(k in name for k in kws):
+                        chosen = f
+                        break
+            if chosen is None and files:
+                chosen = files[0]
+        if chosen is not None and float(total_duration or 0.0) > 0:
+            bgm_track.clips.append(TimelineClip(
+                id="bgm-0",
+                type="bgm",
+                start=0.0,
+                duration=float(total_duration),
+                source_path=str(chosen),
+                source_url=f"/static/assets/bgm/{chosen.name}",
+                volume=0.25,
+                style={"loop": True},
+            ))
+    except Exception:
+        pass
+    tracks.append(bgm_track)
+
+    # 6. 贴纸轨道
+    sticker_track = Track(
+        id="sticker-main",
+        name="贴纸",
+        type="sticker",
+        clips=[],
+        muted=False
+    )
+    tracks.append(sticker_track)
+    
+    return EditorState(
+        project_id=project_id,
+        total_duration=total_duration,
+        tracks=tracks,
+        current_time=0,
+        zoom=1.0
+    )
+
+
+@router.post("/{project_id}/state")
+async def save_editor_state(project_id: str, state: EditorState):
+    """保存编辑器状态到数据库"""
+    project = db.get_project(project_id)
+    if not project:
+        raise HTTPException(status_code=404, detail="项目不存在")
+
+    # 1) 持久化 editor_state（用于 Cut/Trim/Split 回放，不改表结构）
+    product_info = project.get("product_info", {}) or {}
+    product_info["editor_state"] = state.model_dump()
+    db.update_project_product_info(project_id, product_info)
+    
+    # 将编辑器状态转换回 script_data 格式
+    script_data = project.get("script_data", {})
+    
+    # 更新 voiceover_timeline
+    voiceover_timeline = []
+    subtitle_clips = []
+    
+    for track in state.tracks:
+        if track.type == "audio" and track.id == "audio-voiceover":
+            for clip in track.clips:
+                voiceover_timeline.append({
+                    "text": clip.text or "",
+                    "start_time": clip.start,
+                    "duration": clip.duration
+                })
+        elif track.type == "subtitle":
+            for clip in track.clips:
+                subtitle_clips.append({
+                    "text": clip.text or "",
+                    "subtitle": clip.text or "",
+                    "start_time": clip.start,
+                    "duration": clip.duration
+                })
+    
+    # 合并字幕到 voiceover_timeline
+    for i, vo in enumerate(voiceover_timeline):
+        if i < len(subtitle_clips):
+            vo["subtitle"] = subtitle_clips[i].get("text", vo.get("text", ""))
+    
+    script_data["voiceover_timeline"] = voiceover_timeline
+    
+    # 更新花字
+    for track in state.tracks:
+        if track.type == "fancy_text":
+            for clip in track.clips:
+                # 找到对应的 scene
+                scene_id_str = clip.id.replace("fancy-", "")
+                try:
+                    scene_id = int(scene_id_str)
+                    for scene in script_data.get("scenes", []):
+                        if scene["id"] == scene_id:
+                            if "fancy_text" not in scene:
+                                scene["fancy_text"] = {}
+                            scene["fancy_text"]["text"] = clip.text or ""
+                            scene["fancy_text"]["start_time"] = clip.start
+                            scene["fancy_text"]["duration"] = clip.duration
+                            break
+                except ValueError:
+                    pass
+    
+    db.update_project_script(project_id, script_data)
+    
+    return {"message": "编辑器状态已保存"}
+
+
+@router.post("/generate-voiceover")
+async def generate_voiceover(request: VoiceoverRequest):
+    """
+    生成 TTS 音频
+    返回音频文件路径
+    """
+    try:
+        output_path = str(config.TEMP_DIR / f"vo_{int(time.time())}.mp3")
+        
+        audio_path = factory.generate_voiceover_volcengine(
+            text=request.text,
+            voice_type=request.voice_type,
+            output_path=output_path
+        )
+        
+        if audio_path and os.path.exists(audio_path):
+            # 如果需要调整时长
+            if request.target_duration:
+                adjusted_path = str(config.TEMP_DIR / f"vo_adj_{int(time.time())}.mp3")
+                ffmpeg_utils.fit_audio_to_duration_by_speed(audio_path, request.target_duration, adjusted_path)
+                audio_path = adjusted_path
+            
+            # 返回源时长（用于前端计算倍速）
+            try:
+                dur = float(ffmpeg_utils.get_audio_info(audio_path).get("duration") or 0.0)
+            except Exception:
+                dur = 0.0
+            return {
+                "success": True,
+                "path": audio_path,
+                "url": f"/static/temp/{Path(audio_path).name}",
+                "duration": dur if dur > 0 else None,
+            }
+        else:
+            raise HTTPException(status_code=500, detail="TTS 生成失败")
+            
+    except Exception as e:
+        logger.error(f"TTS 生成错误: {e}")
+        raise HTTPException(status_code=500, detail=str(e))
+
+
+@router.post("/generate-fancy-text")
+async def generate_fancy_text(request: FancyTextRequest):
+    """
+    生成花字图片
+    返回图片路径
+    """
+    try:
+        img_path = renderer.render(
+            text=request.text,
+            style=request.style,
+            cache=False
+        )
+        
+        if img_path and os.path.exists(img_path):
+            return {
+                "success": True,
+                "path": img_path,
+                "url": f"/static/temp/{Path(img_path).name}"
+            }
+        else:
+            raise HTTPException(status_code=500, detail="花字生成失败")
+            
+    except Exception as e:
+        logger.error(f"花字生成错误: {e}")
+        raise HTTPException(status_code=500, detail=str(e))
+
+
+@router.post("/trim-video")
+async def trim_video(request: TrimRequest):
+    """
+    裁剪视频片段
+    返回新视频路径
+    """
+    if not os.path.exists(request.source_path):
+        raise HTTPException(status_code=404, detail="源视频不存在")
+    
+    try:
+        output_path = str(config.TEMP_DIR / f"trimmed_{int(time.time())}.mp4")
+        
+        # 使用 ffmpeg 裁剪
+        duration = request.end_time - request.start_time
+        cmd = [
+            ffmpeg_utils.FFMPEG_PATH, "-y",
+            "-ss", str(request.start_time),
+            "-i", request.source_path,
+            "-t", str(duration),
+            "-c", "copy",
+            output_path
+        ]
+        ffmpeg_utils._run_ffmpeg(cmd)
+        
+        return {
+            "success": True,
+            "path": output_path,
+            "url": f"/static/temp/{Path(output_path).name}",
+            "duration": duration
+        }
+        
+    except Exception as e:
+        logger.error(f"视频裁剪错误: {e}")
+        raise HTTPException(status_code=500, detail=str(e))
+
+
+@router.delete("/{project_id}/clip/{clip_id}")
+async def delete_clip(project_id: str, clip_id: str):
+    """删除时间轴上的片段"""
+    # 这里主要是前端状态管理，后端只做记录
+    logger.info(f"删除片段: {project_id}/{clip_id}")
+    return {"message": "片段已删除", "clip_id": clip_id}
+
+
+

api/routes/projects.py

@@ -0,0 +1,315 @@
+"""
+项目管理 API 路由
+提供项目 CRUD 和状态查询
+"""
+import os
+import time
+import logging
+from typing import List, Optional, Any, Dict
+from pathlib import Path
+
+from fastapi import APIRouter, HTTPException, UploadFile, File, Form
+from pydantic import BaseModel
+
+import config
+from modules.db_manager import db
+from modules.script_gen import ScriptGenerator
+from modules.image_gen import ImageGenerator
+from modules.video_gen import VideoGenerator
+from modules.legacy_path_mapper import map_legacy_local_path
+
+logger = logging.getLogger(__name__)
+router = APIRouter()
+
+
+# ============================================================
+# Pydantic Models
+# ============================================================
+
+class ProductInfo(BaseModel):
+    category: str = ""
+    price: str = ""
+    tags: str = ""
+    params: str = ""
+    style_hint: str = ""
+
+
+class ProjectCreateRequest(BaseModel):
+    name: str
+    product_info: ProductInfo
+
+
+class ProjectResponse(BaseModel):
+    id: str
+    name: str
+    status: str
+    product_info: dict
+    script_data: Optional[dict] = None
+    created_at: float
+    updated_at: float
+
+
+class ProjectListItem(BaseModel):
+    id: str
+    name: str
+    status: str
+    updated_at: float
+
+
+class PromptDebugResponse(BaseModel):
+    """
+    导出当次脚本生成实际使用的 prompt（来自 projects.script_data._debug）。
+    注意：默认不返回 raw_output，避免体积过大/泄露无关信息。
+    """
+    project_id: str
+    provider: Optional[str] = None
+    system_prompt: Optional[str] = None
+    user_prompt: Optional[str] = None
+    image_urls: Optional[List[str]] = None
+    raw_output: Optional[str] = None
+
+
+class SceneAssetResponse(BaseModel):
+    scene_id: int
+    asset_type: str
+    status: str
+    local_path: Optional[str] = None
+    url: Optional[str] = None
+
+
+# ============================================================
+# API Endpoints
+# ============================================================
+
+@router.get("", response_model=List[ProjectListItem])
+async def list_projects():
+    """获取所有项目列表"""
+    projects = db.list_projects()
+    return projects
+
+
+@router.post("", response_model=dict)
+async def create_project(request: ProjectCreateRequest):
+    """创建新项目"""
+    project_id = f"PROJ-{int(time.time())}"
+    
+    product_info_dict = request.product_info.model_dump()
+    db.create_project(project_id, request.name, product_info_dict)
+    
+    return {
+        "id": project_id,
+        "message": "项目创建成功"
+    }
+
+
+@router.get("/{project_id}", response_model=ProjectResponse)
+async def get_project(project_id: str):
+    """获取项目详情"""
+    project = db.get_project(project_id)
+    if not project:
+        raise HTTPException(status_code=404, detail="项目不存在")
+    return project
+
+
+@router.get("/{project_id}/prompt-debug", response_model=PromptDebugResponse)
+async def get_prompt_debug(project_id: str, include_raw_output: bool = False):
+    """
+    获取该项目“生成脚本时实际使用”的 system_prompt / user_prompt。
+    - 数据来源：projects.script_data._debug（由 ScriptGenerator 在生成脚本时写入）
+    - include_raw_output=true 时返回 raw_output（可能很大）
+    """
+    project = db.get_project(project_id)
+    if not project:
+        raise HTTPException(status_code=404, detail="项目不存在")
+
+    script_data: Dict[str, Any] = project.get("script_data") or {}
+    debug: Dict[str, Any] = script_data.get("_debug") or {}
+
+    resp = {
+        "project_id": project_id,
+        "provider": debug.get("provider"),
+        "system_prompt": debug.get("system_prompt"),
+        "user_prompt": debug.get("user_prompt"),
+        "image_urls": debug.get("image_urls"),
+    }
+    if include_raw_output:
+        resp["raw_output"] = debug.get("raw_output")
+    return resp
+
+
+@router.get("/{project_id}/assets", response_model=List[dict])
+async def get_project_assets(project_id: str):
+    """获取项目所有素材"""
+    assets = db.get_assets(project_id)
+    
+    # 添加可访问的 URL
+    for asset in assets:
+        source_path, mapped_url = map_legacy_local_path(asset.get("local_path"))
+        if source_path and os.path.exists(source_path):
+            # 转换为相对路径 URL
+            local_path = Path(source_path)
+            if mapped_url:
+                asset["url"] = mapped_url
+                asset["local_path"] = source_path
+            elif str(config.OUTPUT_DIR) in str(local_path):
+                asset["url"] = f"/static/output/{local_path.name}"
+                asset["local_path"] = source_path
+            elif str(config.TEMP_DIR) in str(local_path):
+                asset["url"] = f"/static/temp/{local_path.name}"
+                asset["local_path"] = source_path
+    
+    return assets
+
+
+@router.post("/{project_id}/upload-images")
+async def upload_product_images(
+    project_id: str,
+    files: List[UploadFile] = File(...)
+):
+    """上传商品主图"""
+    project = db.get_project(project_id)
+    if not project:
+        raise HTTPException(status_code=404, detail="项目不存在")
+    
+    saved_paths = []
+    for file in files:
+        # 保存到 temp 目录
+        file_path = config.TEMP_DIR / file.filename
+        with open(file_path, "wb") as f:
+            content = await file.read()
+            f.write(content)
+        saved_paths.append(str(file_path))
+    
+    # 更新项目 product_info
+    product_info = project.get("product_info", {})
+    product_info["uploaded_images"] = saved_paths
+    
+    # 注意：这里需要重新保存整个项目，简化处理
+    # 实际应该添加一个 update_product_info 方法
+    
+    return {
+        "message": f"上传成功 {len(saved_paths)} 张图片",
+        "paths": saved_paths
+    }
+
+
+@router.post("/{project_id}/generate-script")
+async def generate_script(
+    project_id: str,
+    model_provider: str = "shubiaobiao"
+):
+    """生成脚本"""
+    project = db.get_project(project_id)
+    if not project:
+        raise HTTPException(status_code=404, detail="项目不存在")
+    
+    product_info = project.get("product_info", {})
+    image_paths = product_info.get("uploaded_images", [])
+    
+    gen = ScriptGenerator()
+    script = gen.generate_script(
+        project["name"],
+        product_info,
+        image_paths,
+        model_provider=model_provider
+    )
+    
+    if script:
+        db.update_project_script(project_id, script)
+        return {
+            "message": "脚本生成成功",
+            "script": script
+        }
+    else:
+        raise HTTPException(status_code=500, detail="脚本生成失败")
+
+
+@router.post("/{project_id}/generate-images")
+async def generate_images(
+    project_id: str,
+    model_provider: str = "shubiaobiao"
+):
+    """生成分镜图片"""
+    project = db.get_project(project_id)
+    if not project:
+        raise HTTPException(status_code=404, detail="项目不存在")
+    
+    script_data = project.get("script_data")
+    if not script_data:
+        raise HTTPException(status_code=400, detail="请先生成脚本")
+    
+    product_info = project.get("product_info", {})
+    base_imgs = product_info.get("uploaded_images", [])
+    
+    if not base_imgs:
+        raise HTTPException(status_code=400, detail="请先上传商品主图")
+    
+    img_gen = ImageGenerator()
+    scenes = script_data.get("scenes", [])
+    visual_anchor = script_data.get("visual_anchor", "")
+    
+    results = {}
+    current_refs = list(base_imgs)
+    
+    for scene in scenes:
+        scene_id = scene["id"]
+        img_path = img_gen.generate_single_scene_image(
+            scene=scene,
+            original_image_path=current_refs,
+            previous_image_path=None,
+            model_provider=model_provider,
+            visual_anchor=visual_anchor
+        )
+        
+        if img_path:
+            results[scene_id] = img_path
+            current_refs.append(img_path)
+            db.save_asset(project_id, scene_id, "image", "completed", local_path=img_path)
+    
+    db.update_project_status(project_id, "images_generated")
+    
+    return {
+        "message": f"生成成功 {len(results)} 张图片",
+        "images": results
+    }
+
+
+@router.post("/{project_id}/generate-videos")
+async def generate_videos(project_id: str):
+    """生成分镜视频"""
+    project = db.get_project(project_id)
+    if not project:
+        raise HTTPException(status_code=404, detail="项目不存在")
+    
+    script_data = project.get("script_data")
+    if not script_data:
+        raise HTTPException(status_code=400, detail="请先生成脚本")
+    
+    # 获取已生成的图片
+    assets = db.get_assets(project_id, "image")
+    scene_images = {a["scene_id"]: a["local_path"] for a in assets if a["status"] == "completed"}
+    
+    if not scene_images:
+        raise HTTPException(status_code=400, detail="请先生成分镜图片")
+    
+    vid_gen = VideoGenerator()
+    videos = vid_gen.generate_scene_videos(
+        project_id,
+        script_data,
+        scene_images
+    )
+    
+    if videos:
+        for sid, path in videos.items():
+            db.save_asset(project_id, sid, "video", "completed", local_path=path)
+        db.update_project_status(project_id, "videos_generated")
+        
+        return {
+            "message": f"生成成功 {len(videos)} 个视频",
+            "videos": videos
+        }
+    else:
+        raise HTTPException(status_code=500, detail="视频生成失败")
+
+
+

api/tasks/__init__.py

@@ -0,0 +1,15 @@
+# Celery Tasks Package
+
+
+
+
+
+
+
+
+
+
+
+
+
+

api/tasks/audio_tasks.py

@@ -0,0 +1,212 @@
+"""
+音频处理 Celery 任务
+TTS 生成、花字渲染等
+"""
+import os
+import time
+import logging
+from pathlib import Path
+from typing import Dict, Any, Optional
+
+from celery import shared_task
+
+import config
+from modules import factory, ffmpeg_utils
+from modules.text_renderer import renderer
+
+logger = logging.getLogger(__name__)
+
+
+@shared_task(bind=True, name="audio.generate_tts")
+def generate_tts_task(
+    self,
+    text: str,
+    voice_type: str = "zh_female_santongyongns_saturn_bigtts",
+    target_duration: Optional[float] = None,
+    output_path: Optional[str] = None
+) -> Dict[str, Any]:
+    """
+    生成 TTS 音频（异步任务）
+    
+    Args:
+        text: 要转换的文本
+        voice_type: TTS 音色
+        target_duration: 目标时长（秒），如果指定会调整音频速度
+        output_path: 输出路径
+    
+    Returns:
+        {"status": "success", "path": "...", "url": "..."}
+    """
+    task_id = self.request.id
+    logger.info(f"[Task {task_id}] 生成 TTS: {text[:30]}...")
+    
+    if not output_path:
+        timestamp = int(time.time())
+        output_path = str(config.TEMP_DIR / f"tts_{timestamp}.mp3")
+    
+    try:
+        # 生成 TTS
+        audio_path = factory.generate_voiceover_volcengine(
+            text=text,
+            voice_type=voice_type,
+            output_path=output_path
+        )
+        
+        if not audio_path or not os.path.exists(audio_path):
+            raise RuntimeError("TTS 生成失败")
+        
+        # 如果需要调整时长
+        if target_duration:
+            adjusted_path = str(config.TEMP_DIR / f"tts_adj_{int(time.time())}.mp3")
+            ffmpeg_utils.adjust_audio_duration(audio_path, target_duration, adjusted_path)
+            
+            # 删除原始文件
+            if audio_path != output_path:
+                os.remove(audio_path)
+            
+            audio_path = adjusted_path
+        
+        output_url = f"/static/temp/{Path(audio_path).name}"
+        
+        return {
+            "status": "success",
+            "path": audio_path,
+            "url": output_url,
+            "task_id": task_id
+        }
+        
+    except Exception as e:
+        logger.error(f"[Task {task_id}] TTS 生成失败: {e}")
+        raise
+
+
+@shared_task(bind=True, name="audio.generate_fancy_text")
+def generate_fancy_text_task(
+    self,
+    text: str,
+    style: Optional[Dict[str, Any]] = None
+) -> Dict[str, Any]:
+    """
+    生成花字图片（异步任务）
+    
+    Args:
+        text: 花字文本
+        style: 样式配置
+    
+    Returns:
+        {"status": "success", "path": "...", "url": "..."}
+    """
+    task_id = self.request.id
+    logger.info(f"[Task {task_id}] 生成花字: {text}")
+    
+    if not style:
+        style = {
+            "font_size": 72,
+            "font_color": "#FFFFFF",
+            "stroke": {"color": "#000000", "width": 5}
+        }
+    
+    try:
+        img_path = renderer.render(
+            text=text,
+            style=style,
+            cache=False
+        )
+        
+        if not img_path or not os.path.exists(img_path):
+            raise RuntimeError("花字生成失败")
+        
+        output_url = f"/static/temp/{Path(img_path).name}"
+        
+        return {
+            "status": "success",
+            "path": img_path,
+            "url": output_url,
+            "task_id": task_id
+        }
+        
+    except Exception as e:
+        logger.error(f"[Task {task_id}] 花字生成失败: {e}")
+        raise
+
+
+@shared_task(bind=True, name="audio.batch_generate_tts")
+def batch_generate_tts_task(
+    self,
+    items: list,
+    voice_type: str = "zh_female_santongyongns_saturn_bigtts"
+) -> Dict[str, Any]:
+    """
+    批量生成 TTS 音频（异步任务）
+    
+    Args:
+        items: [{"text": "...", "target_duration": 3.0}, ...]
+        voice_type: TTS 音色
+    
+    Returns:
+        {"status": "success", "results": [...]}
+    """
+    task_id = self.request.id
+    logger.info(f"[Task {task_id}] 批量生成 TTS: {len(items)} 条")
+    
+    results = []
+    timestamp = int(time.time())
+    
+    for i, item in enumerate(items):
+        text = item.get("text", "")
+        target_duration = item.get("target_duration")
+        
+        if not text:
+            results.append({"index": i, "status": "skipped", "reason": "空文本"})
+            continue
+        
+        try:
+            output_path = str(config.TEMP_DIR / f"tts_batch_{timestamp}_{i}.mp3")
+            
+            audio_path = factory.generate_voiceover_volcengine(
+                text=text,
+                voice_type=voice_type,
+                output_path=output_path
+            )
+            
+            if target_duration and audio_path:
+                adjusted_path = str(config.TEMP_DIR / f"tts_batch_adj_{timestamp}_{i}.mp3")
+                ffmpeg_utils.adjust_audio_duration(audio_path, target_duration, adjusted_path)
+                audio_path = adjusted_path
+            
+            if audio_path:
+                results.append({
+                    "index": i,
+                    "status": "success",
+                    "path": audio_path,
+                    "url": f"/static/temp/{Path(audio_path).name}"
+                })
+            else:
+                results.append({"index": i, "status": "failed", "reason": "生成失败"})
+                
+        except Exception as e:
+            results.append({"index": i, "status": "failed", "reason": str(e)})
+        
+        # 更新进度
+        progress = (i + 1) / len(items)
+        self.update_state(state="PROGRESS", meta={"progress": progress, "completed": i + 1, "total": len(items)})
+    
+    return {
+        "status": "success",
+        "results": results,
+        "task_id": task_id
+    }
+
+
+
+
+
+
+
+
+
+
+
+
+
+

api/tasks/video_tasks.py

@@ -0,0 +1,418 @@
+"""
+视频处理 Celery 任务
+封装现有的 FFmpeg 处理逻辑为异步任务
+"""
+import os
+import time
+import logging
+import shutil
+from pathlib import Path
+from typing import Dict, Any, List, Optional
+
+from celery import shared_task
+
+import config
+from modules.db_manager import db
+from modules.composer import VideoComposer
+from modules import ffmpeg_utils, factory
+from modules.text_renderer import renderer
+
+logger = logging.getLogger(__name__)
+
+
+@shared_task(bind=True, name="video.compose_from_script")
+def compose_from_script_task(
+    self,
+    project_id: str,
+    script_data: Dict[str, Any],
+    video_map: Dict[int, str],
+    bgm_path: Optional[str] = None,
+    voice_type: str = "zh_female_santongyongns_saturn_bigtts",
+    output_name: Optional[str] = None
+) -> Dict[str, Any]:
+    """
+    基于脚本合成视频（异步任务）
+    
+    Args:
+        project_id: 项目 ID
+        script_data: 脚本数据
+        video_map: 场景视频映射
+        bgm_path: BGM 路径
+        voice_type: TTS 音色
+        output_name: 输出文件名
+    
+    Returns:
+        {"status": "success", "output_path": "...", "output_url": "..."}
+    """
+    task_id = self.request.id
+    logger.info(f"[Task {task_id}] 开始合成视频: {project_id}")
+    
+    # 更新任务状态
+    self.update_state(state="PROGRESS", meta={"progress": 0.1, "message": "准备素材..."})
+    
+    try:
+        # 验证视频文件存在
+        valid_videos = {}
+        for scene_id, path in video_map.items():
+            if path and os.path.exists(path):
+                valid_videos[int(scene_id)] = path
+        
+        if not valid_videos:
+            raise ValueError("没有可用的视频素材")
+        
+        self.update_state(state="PROGRESS", meta={"progress": 0.2, "message": "创建合成器..."})
+        
+        # 创建合成器
+        composer = VideoComposer(voice_type=voice_type)
+        
+        # 生成输出名称
+        if not output_name:
+            output_name = f"final_{project_id}_{int(time.time())}"
+        
+        self.update_state(state="PROGRESS", meta={"progress": 0.3, "message": "执行合成..."})
+        
+        # 执行合成
+        output_path = composer.compose_from_script(
+            script=script_data,
+            video_map=valid_videos,
+            bgm_path=bgm_path,
+            output_name=output_name
+        )
+        
+        self.update_state(state="PROGRESS", meta={"progress": 0.9, "message": "保存结果..."})
+        
+        # 更新数据库
+        db.save_asset(project_id, 0, "final_video", "completed", local_path=output_path)
+        db.update_project_status(project_id, "completed")
+        
+        output_url = f"/static/output/{Path(output_path).name}"
+        
+        logger.info(f"[Task {task_id}] 合成完成: {output_path}")
+        
+        return {
+            "status": "success",
+            "output_path": output_path,
+            "output_url": output_url,
+            "task_id": task_id
+        }
+        
+    except Exception as e:
+        logger.error(f"[Task {task_id}] 合成失败: {e}")
+        db.update_project_status(project_id, "failed")
+        raise
+
+
+@shared_task(bind=True, name="video.compose_from_tracks")
+def compose_from_tracks_task(
+    self,
+    project_id: str,
+    video_clips: List[Dict[str, Any]],
+    voiceover_clips: List[Dict[str, Any]],
+    subtitle_clips: List[Dict[str, Any]],
+    fancy_text_clips: List[Dict[str, Any]],
+    bgm_clip: Optional[Dict[str, Any]] = None,
+    voice_type: str = "zh_female_santongyongns_saturn_bigtts",
+    bgm_volume: float = 0.15,
+    output_name: Optional[str] = None
+) -> Dict[str, Any]:
+    """
+    基于编辑器轨道数据合成视频（异步任务）
+    这是核心的多轨合成逻辑
+    """
+    task_id = self.request.id
+    logger.info(f"[Task {task_id}] 开始多轨合成: {project_id}")
+    
+    timestamp = int(time.time())
+    if not output_name:
+        output_name = f"composed_{project_id}_{timestamp}"
+    
+    temp_files = []
+    
+    try:
+        self.update_state(state="PROGRESS", meta={"progress": 0.05, "message": "验证素材..."})
+        
+        # 1. 收集并验证视频片段
+        video_clips = sorted(video_clips, key=lambda x: x.get("start", 0))
+        if not video_clips:
+            raise ValueError("没有视频片段")
+        
+        video_paths = []
+        for clip in video_clips:
+            source_path = clip.get("source_path")
+            if not source_path or not os.path.exists(source_path):
+                continue
+            
+            trim_start = clip.get("trim_start", 0)
+            trim_end = clip.get("trim_end")
+            
+            if trim_start > 0 or trim_end:
+                # 需要裁剪
+                trimmed_path = str(config.TEMP_DIR / f"trim_{timestamp}_{len(video_paths)}.mp4")
+                duration = (trim_end or 999) - trim_start
+                cmd = [
+                    ffmpeg_utils.FFMPEG_PATH, "-y",
+                    "-ss", str(trim_start),
+                    "-i", source_path,
+                    "-t", str(duration),
+                    "-c", "copy",
+                    trimmed_path
+                ]
+                ffmpeg_utils._run_ffmpeg(cmd)
+                video_paths.append(trimmed_path)
+                temp_files.append(trimmed_path)
+            else:
+                video_paths.append(source_path)
+        
+        if not video_paths:
+            raise ValueError("没有可用的视频片段")
+        
+        self.update_state(state="PROGRESS", meta={"progress": 0.15, "message": "拼接视频..."})
+        
+        # 2. 拼接视频
+        merged_path = str(config.TEMP_DIR / f"{output_name}_merged.mp4")
+        ffmpeg_utils.concat_videos(video_paths, merged_path, (1080, 1920))
+        temp_files.append(merged_path)
+        current_video = merged_path
+        
+        # 添加静音轨
+        silent_path = str(config.TEMP_DIR / f"{output_name}_silent.mp4")
+        ffmpeg_utils.add_silence_audio(current_video, silent_path)
+        temp_files.append(silent_path)
+        current_video = silent_path
+        
+        # 获取总时长
+        info = ffmpeg_utils.get_video_info(current_video)
+        total_duration = float(info.get("duration", 10))
+        
+        self.update_state(state="PROGRESS", meta={"progress": 0.3, "message": "生成旁白..."})
+        
+        # 3. 生成并混入旁白
+        if voiceover_clips:
+            mixed_audio_path = str(config.TEMP_DIR / f"{output_name}_mixed_vo.mp3")
+            
+            # 初始化静音底轨
+            ffmpeg_utils._run_ffmpeg([
+                ffmpeg_utils.FFMPEG_PATH, "-y",
+                "-f", "lavfi", "-i", "anullsrc=r=44100:cl=stereo",
+                "-t", str(total_duration),
+                "-c:a", "mp3",
+                mixed_audio_path
+            ])
+            temp_files.append(mixed_audio_path)
+            
+            for i, clip in enumerate(voiceover_clips):
+                text = clip.get("text", "")
+                if not text:
+                    continue
+                
+                start_time = clip.get("start", 0)
+                target_duration = clip.get("duration", 3)
+                
+                # 生成 TTS
+                tts_path = factory.generate_voiceover_volcengine(
+                    text=text,
+                    voice_type=voice_type,
+                    output_path=str(config.TEMP_DIR / f"{output_name}_vo_{i}.mp3")
+                )
+                
+                if not tts_path:
+                    continue
+                temp_files.append(tts_path)
+                
+                # 调整时长
+                adjusted_path = str(config.TEMP_DIR / f"{output_name}_vo_adj_{i}.mp3")
+                ffmpeg_utils.adjust_audio_duration(tts_path, target_duration, adjusted_path)
+                temp_files.append(adjusted_path)
+                
+                # 混合
+                new_mixed = str(config.TEMP_DIR / f"{output_name}_mixed_{i}.mp3")
+                ffmpeg_utils.mix_audio_at_offset(mixed_audio_path, adjusted_path, start_time, new_mixed)
+                mixed_audio_path = new_mixed
+                temp_files.append(new_mixed)
+            
+            # 混入视频
+            voiced_path = str(config.TEMP_DIR / f"{output_name}_voiced.mp4")
+            ffmpeg_utils.mix_audio(
+                current_video, mixed_audio_path, voiced_path,
+                audio_volume=1.5,
+                video_volume=0.2
+            )
+            temp_files.append(voiced_path)
+            current_video = voiced_path
+        
+        self.update_state(state="PROGRESS", meta={"progress": 0.5, "message": "添加字幕..."})
+        
+        # 4. 添加字幕
+        if subtitle_clips:
+            subtitles = []
+            for clip in subtitle_clips:
+                text = clip.get("text", "")
+                if text:
+                    subtitles.append({
+                        "text": ffmpeg_utils.wrap_text_smart(text),
+                        "start": clip.get("start", 0),
+                        "duration": clip.get("duration", 3),
+                        "style": clip.get("style", {})
+                    })
+            
+            if subtitles:
+                subtitled_path = str(config.TEMP_DIR / f"{output_name}_subtitled.mp4")
+                subtitle_style = {
+                    "font": ffmpeg_utils._get_font_path(),
+                    "fontsize": 60,
+                    "fontcolor": "white",
+                    "borderw": 5,
+                    "bordercolor": "black",
+                    "box": 0,
+                    "x": "(w-text_w)/2",
+                    "y": "h-200",
+                }
+                ffmpeg_utils.add_multiple_subtitles(
+                    current_video, subtitles, subtitled_path, default_style=subtitle_style
+                )
+                temp_files.append(subtitled_path)
+                current_video = subtitled_path
+        
+        self.update_state(state="PROGRESS", meta={"progress": 0.65, "message": "叠加花字..."})
+        
+        # 5. 叠加花字
+        if fancy_text_clips:
+            overlay_configs = []
+            for clip in fancy_text_clips:
+                text = clip.get("text", "")
+                if not text:
+                    continue
+                
+                style = clip.get("style", {
+                    "font_size": 72,
+                    "font_color": "#FFFFFF",
+                    "stroke": {"color": "#000000", "width": 5}
+                })
+                
+                img_path = renderer.render(text, style, cache=False)
+                temp_files.append(img_path)
+                
+                position = clip.get("position", {})
+                overlay_configs.append({
+                    "path": img_path,
+                    "x": position.get("x", "(W-w)/2"),
+                    "y": position.get("y", "180"),
+                    "start": clip.get("start", 0),
+                    "duration": clip.get("duration", 5)
+                })
+            
+            if overlay_configs:
+                fancy_path = str(config.TEMP_DIR / f"{output_name}_fancy.mp4")
+                ffmpeg_utils.overlay_multiple_images(current_video, overlay_configs, fancy_path)
+                temp_files.append(fancy_path)
+                current_video = fancy_path
+        
+        self.update_state(state="PROGRESS", meta={"progress": 0.8, "message": "添加背景音乐..."})
+        
+        # 6. 添加 BGM
+        if bgm_clip:
+            bgm_source = bgm_clip.get("source_path")
+            if bgm_source and os.path.exists(bgm_source):
+                bgm_output = str(config.TEMP_DIR / f"{output_name}_bgm.mp4")
+                ffmpeg_utils.add_bgm(
+                    current_video, bgm_source, bgm_output,
+                    bgm_volume=bgm_volume
+                )
+                temp_files.append(bgm_output)
+                current_video = bgm_output
+        
+        self.update_state(state="PROGRESS", meta={"progress": 0.9, "message": "保存输出..."})
+        
+        # 7. 输出最终文件
+        final_path = str(config.OUTPUT_DIR / f"{output_name}.mp4")
+        shutil.copy(current_video, final_path)
+        
+        # 更新数据库
+        db.save_asset(project_id, 0, "final_video", "completed", local_path=final_path)
+        db.update_project_status(project_id, "completed")
+        
+        output_url = f"/static/output/{Path(final_path).name}"
+        
+        logger.info(f"[Task {task_id}] 多轨合成完成: {final_path}")
+        
+        return {
+            "status": "success",
+            "output_path": final_path,
+            "output_url": output_url,
+            "task_id": task_id
+        }
+        
+    except Exception as e:
+        logger.error(f"[Task {task_id}] 多轨合成失败: {e}")
+        db.update_project_status(project_id, "failed")
+        raise
+        
+    finally:
+        # 清理临时文件
+        for f in temp_files:
+            try:
+                if os.path.exists(f):
+                    os.remove(f)
+            except:
+                pass
+
+
+@shared_task(bind=True, name="video.trim")
+def trim_video_task(
+    self,
+    source_path: str,
+    start_time: float,
+    end_time: float,
+    output_path: Optional[str] = None
+) -> Dict[str, Any]:
+    """
+    裁剪视频片段（异步任务）
+    """
+    task_id = self.request.id
+    logger.info(f"[Task {task_id}] 裁剪视频: {source_path}")
+    
+    if not os.path.exists(source_path):
+        raise FileNotFoundError(f"源视频不存在: {source_path}")
+    
+    if not output_path:
+        timestamp = int(time.time())
+        output_path = str(config.TEMP_DIR / f"trimmed_{timestamp}.mp4")
+    
+    try:
+        duration = end_time - start_time
+        cmd = [
+            ffmpeg_utils.FFMPEG_PATH, "-y",
+            "-ss", str(start_time),
+            "-i", source_path,
+            "-t", str(duration),
+            "-c", "copy",
+            output_path
+        ]
+        ffmpeg_utils._run_ffmpeg(cmd)
+        
+        output_url = f"/static/temp/{Path(output_path).name}"
+        
+        return {
+            "status": "success",
+            "output_path": output_path,
+            "output_url": output_url,
+            "duration": duration
+        }
+        
+    except Exception as e:
+        logger.error(f"[Task {task_id}] 裁剪失败: {e}")
+        raise
+
+
+
+
+
+
+
+
+
+
+
+
+
+

app.py

@@ -26,6 +26,7 @@ from modules import path_utils
 from modules import limits
 from modules.legacy_path_mapper import map_legacy_local_path
 from modules.legacy_normalizer import normalize_legacy_project
+import extra_streamlit_components as stx
 
 # Page Config
 st.set_page_config(
@@ -35,6 +36,84 @@ st.set_page_config(
     initial_sidebar_state="expanded"
 )
 
+# ============================================================
+# Auth (login + remember cookie)
+# ============================================================
+COOKIE_NAME = "vf_session"
+cookie_manager = stx.CookieManager(key="vf_cookie_mgr")
+
+def _current_user() -> dict:
+    u = st.session_state.get("current_user")
+    return u if isinstance(u, dict) else None
+
+def _set_current_user(u: dict):
+    st.session_state.current_user = u
+
+def _try_restore_login():
+    if _current_user():
+        return
+    try:
+        token = cookie_manager.get(COOKIE_NAME)
+    except Exception:
+        token = None
+    if token:
+        u = db.validate_session(token)
+        if u:
+            _set_current_user(u)
+
+def _logout():
+    try:
+        token = cookie_manager.get(COOKIE_NAME)
+    except Exception:
+        token = None
+    if token:
+        db.revoke_session(token)
+    try:
+        cookie_manager.delete(COOKIE_NAME)
+    except Exception:
+        pass
+    st.session_state.pop("current_user", None)
+    st.rerun()
+
+def _login_gate():
+    _try_restore_login()
+    u = _current_user()
+    if u:
+        return
+    st.title("登录")
+    # Login page runs before the global CSS block below; inject minimal CSS here.
+    st.markdown(
+        """
+<style>
+button[aria-label="Show password text"],
+button[aria-label="Hide password text"] {
+  display: none !important;
+}
+</style>
+""",
+        unsafe_allow_html=True,
+    )
+    c1, c2 = st.columns([1, 2])
+    with c1:
+        username = st.text_input("用户名", value="", key="login_user")
+        password = st.text_input("密码", value="", type="password", key="login_pass")
+        if st.button("登录", type="primary"):
+            au = db.authenticate_user(username.strip(), password)
+            if not au:
+                st.error("用户名或密码错误，或账号已禁用")
+                st.stop()
+            token = db.create_session(au["id"])
+            try:
+                cookie_manager.set(COOKIE_NAME, token, max_age=7 * 24 * 3600)
+            except Exception:
+                # fallback: session-only
+                pass
+            _set_current_user(au)
+            st.rerun()
+    st.stop()
+
+_login_gate()
+
 # ============================================================
 # BGM 智能匹配函数
 # ============================================================
@@ -109,6 +188,12 @@ st.markdown("""
     .stTextInput input, .stTextArea textarea {
         border-radius: 4px;
     }
+
+    /* Hide Streamlit password reveal (eye) buttons to avoid accidental exposure */
+    button[aria-label="Show password text"],
+    button[aria-label="Hide password text"] {
+        display: none !important;
+    }
 </style>
 """, unsafe_allow_html=True)
 
@@ -150,9 +235,9 @@ def _ui_key(suffix: str) -> str:
 
 def load_project(project_id):
     """Load project state from DB"""
-    data = db.get_project(project_id)
+    data = db.get_project_for_user(project_id, _current_user())
     if not data:
-        st.error("Project not found")
+        st.error("Project not found / no permission")
         return
     
     st.session_state.project_id = project_id
@@ -203,6 +288,7 @@ def load_project(project_id):
         st.session_state.uploaded_images = []
     
     # Restore assets
+    # RBAC: assets are project-scoped, so permission already checked above.
     assets = db.get_assets(project_id)
     images = {}
     videos = {}
@@ -238,11 +324,107 @@ def load_project(project_id):
     else:
         st.session_state.current_step = 0
 
+# ============================================================
+# Helper Functions (must be defined before Sidebar uses them)
+# ============================================================
+def _record_metrics(project_id: str, patch: dict):
+    """Persist lightweight timing/diagnostic metrics into project.product_info['_metrics']."""
+    if not project_id or not isinstance(patch, dict) or not patch:
+        return
+    try:
+        proj = db.get_project(project_id) or {}
+        product_info = proj.get("product_info") or {}
+        metrics = product_info.get("_metrics") if isinstance(product_info.get("_metrics"), dict) else {}
+        metrics.update(patch)
+        metrics["updated_at"] = time.time()
+        product_info["_metrics"] = metrics
+        db.update_project_product_info(project_id, product_info)
+    except Exception:
+        # metrics must never break UX
+        pass
+
+
+def _get_metrics(project_id: str) -> dict:
+    try:
+        proj = db.get_project(project_id) or {}
+        product_info = proj.get("product_info") or {}
+        m = product_info.get("_metrics")
+        return m if isinstance(m, dict) else {}
+    except Exception:
+        return {}
+
+
+def _ensure_local_videos(project_id: str, scenes: list):
+    """
+    Step 5 合成前的保障逻辑：
+    检查分镜视频是否已下载到服务器本地。如果只有 URL 没有本地文件，则后台静默下载。
+    """
+    if not project_id or not scenes:
+        return
+        
+    vid_gen = VideoGenerator()
+    downloaded = 0
+    missing_assets = []
+    
+    for scene in scenes:
+        scene_id = scene["id"]
+        local_path = st.session_state.scene_videos.get(scene_id)
+        
+        # 如果本地文件不存在，尝试补全
+        if not local_path or not os.path.exists(local_path):
+            asset = db.get_asset(project_id, scene_id, "video")
+            if not asset:
+                missing_assets.append(f"Scene {scene_id}")
+                continue
+                
+            meta = asset.get("metadata") or {}
+            video_url = meta.get("video_url")
+            task_id = asset.get("task_id")
+            
+            # 如果 DB 里没有 URL，但有 task_id，尝试现场查询一次
+            if not video_url and task_id:
+                logger.info(f"Checking task {task_id} status for Scene {scene_id} during composition...")
+                status, url = vid_gen.check_task_status(task_id)
+                if status == "succeeded" and url:
+                    video_url = url
+                    db.update_asset_metadata(project_id, scene_id, "video", {"video_url": url, "volc_status": status})
+            
+            if video_url:
+                out_name = path_utils.unique_filename(
+                    prefix="scene_video",
+                    ext="mp4",
+                    project_id=project_id,
+                    scene_id=scene_id,
+                    extra=f"auto_{int(time.time())}"
+                )
+                logger.info(f"Auto-downloading missing video for Scene {scene_id} from {video_url}")
+                target_dir = path_utils.project_videos_dir(project_id)
+                target_path = str(target_dir / out_name)
+                if vid_gen._download_video_to(video_url, target_path):
+                    st.session_state.scene_videos[scene_id] = target_path
+                    db.save_asset(project_id, scene_id, "video", "completed", local_path=target_path)
+                    downloaded += 1
+            else:
+                missing_assets.append(f"Scene {scene_id} (未生成/无URL)")
+    
+    if downloaded > 0:
+        logger.info(f"Successfully auto-downloaded {downloaded} missing videos for project {project_id}")
+    
+    if missing_assets:
+        msg = f"无法合成：以下分镜缺少视频素材，请先在第 4 步生成：{', '.join(missing_assets)}"
+        logger.warning(msg)
+        raise ValueError(msg)
+
 # ============================================================
 # Sidebar
 # ============================================================
 with st.sidebar:
     st.title("📽️ Video Flow")
+    cu = _current_user()
+    if cu:
+        st.caption(f"登录用户: {cu.get('username')} ({cu.get('role')})")
+        if st.button("退出登录", type="secondary"):
+            _logout()
     
     # Mode Selection - 正确计算 index
     mode_options = ["🛠️ 工作台", "📜 历史任务", "⚙️ 设置"]
@@ -262,7 +444,7 @@ with st.sidebar:
     if st.session_state.view_mode == "workspace":
         # Project Selection
         st.subheader("Current Project")
-        projects = db.list_projects()
+        projects = db.list_projects_for_user(_current_user())
         proj_options = {p['id']: f"{p.get('name', 'Untitled')} ({p['id']})" for p in projects}
         
         selected_proj_id = st.selectbox(
@@ -307,13 +489,6 @@ with st.sidebar:
                     for k in keys:
                         if k in m:
                             st.caption(f"{k}: {m.get(k)}")
-        # 在线剪辑入口（React Editor）
-        web_base_url = os.getenv("WEB_BASE_URL", "http://localhost:3000").rstrip("/")
-        st.markdown(
-            f"[打开在线剪辑器]({web_base_url}/editor/{st.session_state.project_id})",
-            unsafe_allow_html=False,
-        )
-        
         st.markdown("---")
         
         # Navigation / Progress
@@ -335,35 +510,6 @@ with st.sidebar:
                 if key != "view_mode": del st.session_state[key]
             st.rerun()
 
-# ============================================================
-# Helper Functions
-# ============================================================
-def _record_metrics(project_id: str, patch: dict):
-    """Persist lightweight timing/diagnostic metrics into project.product_info['_metrics']."""
-    if not project_id or not isinstance(patch, dict) or not patch:
-        return
-    try:
-        proj = db.get_project(project_id) or {}
-        product_info = proj.get("product_info") or {}
-        metrics = product_info.get("_metrics") if isinstance(product_info.get("_metrics"), dict) else {}
-        metrics.update(patch)
-        metrics["updated_at"] = time.time()
-        product_info["_metrics"] = metrics
-        db.update_project_product_info(project_id, product_info)
-    except Exception:
-        # metrics must never break UX
-        pass
-
-
-def _get_metrics(project_id: str) -> dict:
-    try:
-        proj = db.get_project(project_id) or {}
-        product_info = proj.get("product_info") or {}
-        m = product_info.get("_metrics")
-        return m if isinstance(m, dict) else {}
-    except Exception:
-        return {}
-
 def save_uploaded_file(project_id: str, uploaded_file):
     """Save uploaded file to per-project upload dir (avoid overwrites across projects)."""
     if uploaded_file is None:
@@ -467,13 +613,24 @@ if st.session_state.view_mode == "workspace":
             # DB: Create Project
             # 将 uploaded_images 保存到 product_info 以便持久化
             product_info = {"category": category, "price": price, "tags": tags, "params": params, "style_hint": style_hint, "uploaded_images": image_paths}
-            db.create_project(st.session_state.project_id, product_name, product_info)
+            db.create_project(
+                st.session_state.project_id,
+                product_name,
+                product_info,
+                owner_user_id=(_current_user() or {}).get("id"),
+            )
             
             # Call Script Generator
             with st.spinner(f"正在分析商品信息并生成脚本 ({selected_model_label})..."):
                 gen = ScriptGenerator()
                 t0 = perf_counter()
-                script = gen.generate_script(product_name, product_info, image_paths, model_provider=model_provider)
+                script = gen.generate_script(
+                    product_name,
+                    product_info,
+                    image_paths,
+                    model_provider=model_provider,
+                    user_id=(_current_user() or {}).get("id"),
+                )
                 _record_metrics(st.session_state.project_id, {
                     "script_gen_s": round(perf_counter() - t0, 3),
                     "script_model": model_provider,
@@ -689,9 +846,9 @@ if st.session_state.view_mode == "workspace":
                         if not ok:
                             st.warning("系统正在生成其他任务（生图并发已达上限），请稍后再试。")
                             st.stop()
-                        img_gen = ImageGenerator()
+                    img_gen = ImageGenerator()
-                        # Pass ALL uploaded images as reference
+                    # Pass ALL uploaded images as reference
-                        base_imgs = st.session_state.uploaded_images if st.session_state.uploaded_images else []
+                    base_imgs = st.session_state.uploaded_images if st.session_state.uploaded_images else []
                     
                     if not base_imgs: 
                         st.error("No base image found (未找到参考底图). Please upload in Step 1.")
@@ -742,7 +899,7 @@ if st.session_state.view_mode == "workspace":
                         total_scenes = len(scenes)
                         progress_bar = st.progress(0)
                         status_text = st.empty()
-
+                        
                         try:
                             t0 = perf_counter()
                             # Parallel workers within a single run; global semaphore already acquired above.
@@ -755,7 +912,7 @@ if st.session_state.view_mode == "workspace":
                                         img_gen.generate_single_scene_image,
                                         scene=scene,
                                         original_image_path=list(base_imgs),  # ONLY merchant images
-                                        previous_image_path=None,
+                                        previous_image_path=None, 
                                         model_provider=img_provider,
                                         visual_anchor=visual_anchor,
                                         project_id=st.session_state.project_id,
@@ -768,17 +925,15 @@ if st.session_state.view_mode == "workspace":
                                     status_text.text(f"已完成 {done}/{total_scenes}（Scene {scene_id}）")
                                     try:
                                         img_path = fut.result()
+                                        if img_path:
+                                            st.session_state.scene_images[scene_id] = img_path
+                                            db.save_asset(st.session_state.project_id, scene_id, "image", "completed", local_path=img_path)
+                                            # Invalidate stale video for this scene (image changed => old video is wrong)
+                                            db.clear_asset(st.session_state.project_id, scene_id, "video", status="pending")
+                                            st.session_state.scene_videos.pop(scene_id, None)
                                     except Exception as e:
-                                        img_path = None
                                         st.warning(f"Scene {scene_id} 生成失败：{e}")
-
+                                    
-                                    if img_path:
-                                        st.session_state.scene_images[scene_id] = img_path
-                                        db.save_asset(st.session_state.project_id, scene_id, "image", "completed", local_path=img_path)
-                                        # Invalidate stale video for this scene (image changed => old video is wrong)
-                                        db.clear_asset(st.session_state.project_id, scene_id, "video", status="pending")
-                                        st.session_state.scene_videos.pop(scene_id, None)
-
                                     progress_bar.progress(done / total_scenes)
                             
                             status_text.text("生图完成！")
@@ -863,112 +1018,152 @@ if st.session_state.view_mode == "workspace":
             scenes = st.session_state.script_data.get("scenes", [])
             vid_gen = VideoGenerator()
 
-            # Submit-only (non-blocking) to avoid freezing Streamlit under concurrency
+            st.caption("简化策略：第 4 步完成“生成→轮询→下载到服务器本地”。当至少有一个分镜视频落盘后，才允许进入第 5 步合成。")
-            if st.button("🎬 提交图生视频任务（非阻塞）", type="primary"):
+
+            if st.button("🎬 生成分镜视频并下载到服务器（阻塞）", type="primary"):
                 with limits.acquire_video(blocking=False) as ok:
                     if not ok:
                         st.warning("系统正在处理其他视频任务（并发已达上限），请稍后再试。")
                         st.stop()
-                    t0 = perf_counter()
+
-                    submitted = 0
+                    if not st.session_state.project_id:
+                        st.error("缺少 project_id，无法生成视频。")
+                        st.stop()
+
+                    if not scenes:
+                        st.error("脚本中没有分镜，无法生成视频。")
+                        st.stop()
+
+                    # 先把 DB 中已完成且存在的本地视频恢复到 session
                     for scene in scenes:
                         scene_id = scene["id"]
+                        existing = st.session_state.scene_videos.get(scene_id)
+                        if existing and os.path.exists(existing):
+                            continue
+                        asset = db.get_asset(st.session_state.project_id, scene_id, "video")
+                        local_path = (asset or {}).get("local_path")
+                        if local_path and os.path.exists(local_path):
+                            st.session_state.scene_videos[scene_id] = local_path
+
+                    t0 = perf_counter()
+                    total = len(scenes)
+                    done = sum(1 for _sid, p in (st.session_state.scene_videos or {}).items() if p and os.path.exists(p))
+                    progress = st.progress(0.0)
+                    status_text = st.empty()
+
+                    # 收集/提交任务
+                    tasks = {}  # scene_id -> task_id
+                    for scene in scenes:
+                        scene_id = scene["id"]
+                        existing = st.session_state.scene_videos.get(scene_id)
+                        if existing and os.path.exists(existing):
+                            continue
+
+                        asset = db.get_asset(st.session_state.project_id, scene_id, "video")
+                        task_id = (asset or {}).get("task_id")
+                        if task_id:
+                            tasks[scene_id] = task_id
+                            continue
+
                         image_path = st.session_state.scene_images.get(scene_id)
                         prompt = scene.get("video_prompt", "High quality video")
-                        task_id = vid_gen.submit_scene_video_task(
+                        new_task_id = vid_gen.submit_scene_video_task(
                             st.session_state.project_id, scene_id, image_path, prompt
                         )
-                        if task_id:
+                        if new_task_id:
-                            submitted += 1
+                            tasks[scene_id] = new_task_id
-                    _record_metrics(st.session_state.project_id, {
+
-                        "video_submit_s": round(perf_counter() - t0, 3),
+                    if tasks:
-                        "video_submitted": submitted,
+                        db.update_project_status(st.session_state.project_id, "videos_processing")
-                    })
+
-                    if submitted:
+                    out_dir = path_utils.project_videos_dir(st.session_state.project_id)
-                        db.update_project_status(st.session_state.project_id, "videos_processing")
+                    pending = set(tasks.keys())
-                        st.success(f"已提交 {submitted} 个分镜视频任务。可点击下方“刷新恢复”下载结果。")
+                    deadline = time.time() + 15 * 60  # 15 min
-                        time.sleep(0.5)
+
-                        st.rerun()
+                    while pending and time.time() < deadline:
-                    else:
+                        status_text.text(f"视频生成/下载中：已完成 {done}/{total}，队列中 {len(pending)} ...")
-                        st.warning("未提交任何任务（可能缺少图片或接口失败）。")
+                        to_remove = []
+
+                        for scene_id in list(pending):
+                            task_id = tasks.get(scene_id)
+                            if not task_id:
+                                to_remove.append(scene_id)
+                                continue
 
-            if st.button("🔄 刷新状态并恢复已完成任务", type="secondary"):
-                with limits.acquire_video(blocking=False) as ok:
-                    if not ok:
-                        st.warning("系统正在处理其他视频任务（并发已达上限），请稍后再试。")
-                        st.stop()
-                    t0 = perf_counter()
-                    updated = 0
-                    for scene in scenes:
-                        scene_id = scene["id"]
-                        asset = db.get_asset(st.session_state.project_id, scene_id, "video")
-                        if not asset or not asset.get("task_id"):
-                            continue
-                        # if already have local video, skip
-                        existing = st.session_state.scene_videos.get(scene_id)
-                        if existing and os.path.exists(existing):
-                            continue
-                        task_id = asset.get("task_id")
-                        # Query volc status; store URL for direct preview (no server download)
-                        status = None
-                        url = None
-                        # short retries for "succeeded but url missing"
-                        for attempt in range(3):
                             status, url = vid_gen.check_task_status(task_id)
                             if status == "succeeded" and url:
-                                break
+                                out_name = path_utils.unique_filename(
-                            time.sleep(0.5 * (2 ** attempt))
+                                    prefix="scene_video",
+                                    ext="mp4",
+                                    project_id=st.session_state.project_id,
+                                    scene_id=scene_id,
+                                    extra=(task_id[-8:] if isinstance(task_id, str) else None),
+                                )
+                                target_path = str(out_dir / out_name)
+                                ok_dl = vid_gen._download_video_to(url, target_path)
+                                if ok_dl and os.path.exists(target_path):
+                                    st.session_state.scene_videos[scene_id] = target_path
+                                    db.save_asset(
+                                        st.session_state.project_id,
+                                        scene_id,
+                                        "video",
+                                        "completed",
+                                        local_path=target_path,
+                                        task_id=task_id,
+                                        metadata={"downloaded_at": time.time()},
+                                    )
+                                    done += 1
+                                else:
+                                    db.save_asset(
+                                        st.session_state.project_id,
+                                        scene_id,
+                                        "video",
+                                        "failed",
+                                        task_id=task_id,
+                                        metadata={"download_error": True, "checked_at": time.time()},
+                                    )
+                                to_remove.append(scene_id)
+                            elif status in ["failed", "cancelled"]:
+                                db.save_asset(
+                                    st.session_state.project_id,
+                                    scene_id,
+                                    "video",
+                                    "failed",
+                                    task_id=task_id,
+                                    metadata={"volc_status": status, "checked_at": time.time()},
+                                )
+                                to_remove.append(scene_id)
+                            else:
+                                db.update_asset_metadata(
+                                    st.session_state.project_id,
+                                    scene_id,
+                                    "video",
+                                    {"volc_status": status, "checked_at": time.time()},
+                                )
 
-                        meta_patch = {"checked_at": time.time(), "volc_status": status}
+                        for sid in to_remove:
-                        if url:
+                            pending.discard(sid)
-                            meta_patch["video_url"] = url
+
-                        db.update_asset_metadata(st.session_state.project_id, scene_id, "video", meta_patch)
+                        progress.progress(min(1.0, done / max(total, 1)))
-                        updated += 1
+                        if pending:
+                            time.sleep(5)
+
+                    if pending:
+                        st.warning(f"仍有 {len(pending)} 个分镜未在本轮完成：{sorted(list(pending))}。可再次点击按钮继续轮询与下载。")
 
                     _record_metrics(st.session_state.project_id, {
-                        "video_recover_s": round(perf_counter() - t0, 3),
+                        "video_blocking_total_s": round(perf_counter() - t0, 3),
-                        "video_recovered": updated,
+                        "video_done": done,
+                        "video_total": total,
                     })
-                    if updated:
-                        st.success(f"已刷新 {updated} 个分镜状态（成功的将以 URL 直连预览）。")
-                    else:
-                        st.info("暂无可恢复的视频（可能仍在排队/生成中）。")
-                    time.sleep(0.5)
-                    st.rerun()
 
-            if st.button("📥 准备合成素材（下载成功的视频到服务器）", type="secondary"):
+                    if any(p and os.path.exists(p) for p in (st.session_state.scene_videos or {}).values()):
-                with limits.acquire_video(blocking=False) as ok:
+                        db.update_project_status(st.session_state.project_id, "videos_generated")
-                    if not ok:
+                        st.success("已生成并下载到服务器本地。进入第 5 步合成。")
-                        st.warning("系统正在处理其他视频任务（并发已达上限），请稍后再试。")
+                        st.session_state.current_step = 4
-                        st.stop()
+                        st.rerun()
-                    downloaded = 0
-                    for scene in scenes:
-                        scene_id = scene["id"]
-                        existing = st.session_state.scene_videos.get(scene_id)
-                        if existing and os.path.exists(existing):
-                            continue
-                        asset = db.get_asset(st.session_state.project_id, scene_id, "video")
-                        meta = (asset or {}).get("metadata") or {}
-                        video_url = meta.get("video_url")
-                        if not video_url:
-                            continue
-                        out_name = path_utils.unique_filename(
-                            prefix="scene_video",
-                            ext="mp4",
-                            project_id=st.session_state.project_id,
-                            scene_id=scene_id,
-                        )
-                        target_path = str(path_utils.project_videos_dir(st.session_state.project_id) / out_name)
-                        if vid_gen._download_video_to(video_url, target_path):
-                            st.session_state.scene_videos[scene_id] = target_path
-                            db.save_asset(st.session_state.project_id, scene_id, "video", "completed", local_path=target_path, task_id=(asset or {}).get("task_id"), metadata=meta)
-                            downloaded += 1
-                    if downloaded:
-                        st.success(f"已下载 {downloaded} 段视频，可进入合成。")
                     else:
-                        st.info("暂无可下载的视频（请先刷新状态获取 video_url）。")
+                        st.error("未生成任何可用视频，请检查生视频接口或稍后重试。")
-                    time.sleep(0.5)
-                    st.rerun()
             
             # Display Videos (even when partially available)
             if st.session_state.scene_videos or scenes:
@@ -987,43 +1182,14 @@ if st.session_state.view_mode == "workspace":
                         if vid_path and os.path.exists(vid_path):
                             st.video(vid_path)
                         else:
-                            # Try URL preview from DB metadata
                             asset = db.get_asset(st.session_state.project_id, scene_id, "video")
-                            meta = (asset or {}).get("metadata") or {}
+                            status = (asset or {}).get("status") or "pending"
-                            video_url = meta.get("video_url")
+                            task_id = (asset or {}).get("task_id")
-                            if video_url:
+                            if task_id:
-                                # Detect stale mapping: if source image signature differs, warn and avoid misleading preview
+                                st.caption(f"状态: {status} | Task: {str(task_id)[-6:]}")
-                                stale = False
-                                try:
-                                    cur_img = st.session_state.scene_images.get(scene_id)
-                                    if cur_img and os.path.exists(cur_img):
-                                        st_img = stat(cur_img)
-                                        cur_size = int(getattr(st_img, "st_size", 0) or 0)
-                                        cur_mtime = float(getattr(st_img, "st_mtime", 0.0) or 0.0)
-                                        src_size = meta.get("source_image_size")
-                                        src_mtime = meta.get("source_image_mtime")
-                                        if (src_size and cur_size and int(src_size) != cur_size) or (src_mtime and cur_mtime and abs(float(src_mtime) - cur_mtime) > 1e-3):
-                                            stale = True
-                                except Exception:
-                                    stale = False
-                                if stale:
-                                    st.warning("检测到该视频可能基于旧图片生成（图片已更新）。请点击“提交图生视频任务”重新生成，以避免主体不一致。")
-                                st.caption("URL 直连预览（不经服务器落盘）")
-                                st.video(video_url)
                             else:
-                                st.warning("Video missing")
+                                st.caption(f"状态: {status}")
-                            # --- Recovery Logic ---
+                            st.warning("暂无本地视频（请点击上方“生成分镜视频并下载到服务器”）。")
-                            if asset and asset.get("task_id"):
-                                 task_id = asset.get("task_id")
-                                 if st.button(f"🔍 刷新URL (Task {task_id[-6:]})", key=f"recov_{scene_id}"):
-                                     with st.spinner("查询任务状态中..."):
-                                         status, url = vid_gen.check_task_status(task_id)
-                                         patch = {"checked_at": time.time(), "volc_status": status}
-                                         if url:
-                                             patch["video_url"] = url
-                                         db.update_asset_metadata(st.session_state.project_id, scene_id, "video", patch)
-                                         st.success("已刷新任务状态。")
-                                         st.rerun()
                             
                         # Per-scene regenerate button
                         if st.button(f"🔄 重生 S{scene_id}", key=f"regen_vid_{scene_id}"):
@@ -1046,14 +1212,11 @@ if st.session_state.view_mode == "workspace":
                                                     scene_id=scene_id,
                                                     extra=(t_id[-8:] if isinstance(t_id, str) else None),
                                                 )
-                                                new_path = vid_gen._download_video(
+                                                target_dir = path_utils.project_videos_dir(st.session_state.project_id)
-                                                    url,
+                                                target_path = str(target_dir / out_name)
-                                                    out_name,
+                                                if vid_gen._download_video_to(url, target_path) and os.path.exists(target_path):
-                                                    output_dir=path_utils.project_videos_dir(st.session_state.project_id),
+                                                    st.session_state.scene_videos[scene_id] = target_path
-                                                )
+                                                    db.save_asset(st.session_state.project_id, scene_id, "video", "completed", local_path=target_path, task_id=t_id)
-                                                if new_path:
-                                                    st.session_state.scene_videos[scene_id] = new_path
-                                                    db.save_asset(st.session_state.project_id, scene_id, "video", "completed", local_path=new_path, task_id=t_id)
                                                     st.rerun()
                                                 break
                                             elif status in ["failed", "cancelled"]:
@@ -1067,18 +1230,23 @@ if st.session_state.view_mode == "workspace":
                 
                 c_act1, c_act2 = st.columns([1, 4])
                 with c_act1:
-                     if st.button("🔄 重新生成所有视频", type="secondary"):
+                    if st.button("🧹 清空视频并重新生成", type="secondary"):
                         # Clear videos and rerun
                         st.session_state.scene_videos = {}
-                        # Also clear DB video assets to avoid stale URL preview
+                        # Also clear DB video assets
                         if st.session_state.project_id:
                             db.clear_assets(st.session_state.project_id, "video", status="pending")
                         st.rerun()
                         
                 with c_act2:
-                    if st.button("下一步：合成最终成片", type="primary"):
+                    can_compose = any(
+                        p and os.path.exists(p) for p in (st.session_state.scene_videos or {}).values()
+                    )
+                    if st.button("进入第 5 步：合成最终成片", type="primary", disabled=not can_compose):
                         st.session_state.current_step = 4
                         st.rerun()
+                    if not can_compose:
+                        st.caption("⚠️ 需要至少一个分镜视频已下载到服务器本地后，才能进入合成。")
 
     # --- Step 5: Final Composition & Tuning ---
     if st.session_state.current_step >= 4:
@@ -1174,6 +1342,9 @@ if st.session_state.view_mode == "workspace":
                 
                 if st.button("🔄 重新合成 (Re-Compose)", type="primary"):
                     with st.spinner("正在应用修改并重新合成..."):
+                        # 自动补齐视频下载逻辑 (关键优化)
+                        _ensure_local_videos(st.session_state.project_id, st.session_state.script_data.get("scenes", []))
+                        
                         composer = VideoComposer(voice_type=selected_voice)
                         
                         bgm_path = None
@@ -1255,6 +1426,9 @@ if st.session_state.view_mode == "workspace":
                     st.info("暂无合成视频，请先点击开始合成。")
                     if st.button("✨ 开始首次合成", type="primary"):
                         with st.spinner("正在进行多轨合成..."):
+                            # 自动补齐视频下载逻辑 (关键优化)
+                            _ensure_local_videos(st.session_state.project_id, st.session_state.script_data.get("scenes", []))
+                            
                             # Default compose logic with smart BGM matching
                             composer = VideoComposer(voice_type=config.VOLC_TTS_DEFAULT_VOICE)
                             
@@ -1290,6 +1464,9 @@ if st.session_state.view_mode == "workspace":
                 else:
                     st.success(f"共找到 {len(found_files)} 个历史版本")
                     
+                    # 提示用户：如果重新合成，系统会自动补全未下载的素材
+                    st.caption("💡 重新合成将应用当前的微调设置。如果分镜未下载，系统将尝试自动补全。")
+                    
                     # 遍历显示所有历史版本
                     for idx, vid_path in enumerate(found_files):
                         mtime = os.path.getmtime(vid_path)
@@ -1324,7 +1501,7 @@ if st.session_state.view_mode == "workspace":
 elif st.session_state.view_mode == "history":
     st.header("📜 历史任务")
     
-    projects = db.list_projects()
+    projects = db.list_projects_for_user(_current_user())
     
     for proj in projects:
         with st.expander(f"{proj['name']} ({proj['updated_at']})"):
@@ -1394,11 +1571,19 @@ elif st.session_state.view_mode == "settings":
     st.subheader("Prompt 配置")
     
     # Script Generation Prompt
-    current_prompt = db.get_config("prompt_script_gen")
+    cu = _current_user() or {}
+    user_prompt = None
+    try:
+        user_prompt = db.get_user_prompt(cu.get("id"), "prompt_script_gen") if cu.get("id") else None
+    except Exception:
+        user_prompt = None
+    current_prompt = user_prompt or db.get_config("prompt_script_gen")
     
     # 显示当前状态
-    if current_prompt:
+    if user_prompt:
-        st.info("✅ 已加载自定义 Prompt（来自数据库）")
+        st.info("✅ 已加载自定义 Prompt（当前用户）")
+    elif current_prompt:
+        st.info("✅ 已加载自定义 Prompt（全局默认）")
     else:
         st.warning("⚠️ 使用默认 Prompt（数据库中无自定义配置）")
         # Load default from instance if not in DB
@@ -1410,18 +1595,89 @@ elif st.session_state.view_mode == "settings":
     col_save, col_reset = st.columns([1, 3])
     with col_save:
         if st.button("💾 保存配置", type="primary"):
-            db.set_config("prompt_script_gen", new_prompt, "System prompt for script generation step")
+            # Save per-user prompt
-            # 验证保存
+            db.set_user_prompt(cu.get("id"), "prompt_script_gen", new_prompt)
-            saved = db.get_config("prompt_script_gen")
+            saved = db.get_user_prompt(cu.get("id"), "prompt_script_gen")
             if saved == new_prompt:
-                st.success("✅ 配置已保存并验证成功！下次生成脚本时将使用新 Prompt。")
+                st.success("✅ 已保存为“当前用户 Prompt”。下次生成脚本仅影响你自己的账号。")
             else:
                 st.error("❌ 保存可能失败，请检查日志")
     
     with col_reset:
         if st.button("🔄 恢复默认"):
             temp_gen = ScriptGenerator()
-            db.set_config("prompt_script_gen", temp_gen.default_system_prompt, "System prompt for script generation step (DEFAULT)")
+            # Clear per-user override: set empty to remove effect
-            st.success("已恢复默认 Prompt，请刷新页面查看")
+            db.set_user_prompt(cu.get("id"), "prompt_script_gen", "")
+            st.success("已清除当前用户 Prompt，将回退到全局/默认 Prompt。")
+            st.rerun()
+
+    st.markdown("---")
+    st.subheader("账号与权限")
+
+    # Password change for current user
+    with st.expander("修改我的密码", expanded=False):
+        p1 = st.text_input("新密码", type="password", key=_ui_key("pwd_new"))
+        p2 = st.text_input("确认新密码", type="password", key=_ui_key("pwd_new2"))
+        if st.button("保存新密码", type="primary", key=_ui_key("pwd_save")):
+            if not p1 or len(p1) < 6:
+                st.error("密码至少 6 位")
+            elif p1 != p2:
+                st.error("两次输入不一致")
+            else:
+                db.reset_user_password(cu.get("id"), p1)
+                st.success("密码已更新，请重新登录。")
+                _logout()
+
+    # Admin console
+    if (cu.get("role") == "admin"):
+        st.markdown("### Admin：账号管理")
+        users = db.list_users()
+        if users:
+            st.dataframe(
+                [{k: u.get(k) for k in ["username", "role", "is_active", "last_login_at", "created_at", "id"]} for u in users],
+                use_container_width=True,
+            )
+
+        with st.expander("创建/更新用户", expanded=False):
+            u_name = st.text_input("用户名", key=_ui_key("adm_u_name"))
+            u_role = st.selectbox("角色", ["user", "admin"], index=0, key=_ui_key("adm_u_role"))
+            u_active = st.selectbox("是否启用", ["启用", "禁用"], index=0, key=_ui_key("adm_u_active"))
+            u_pwd = st.text_input("初始/重置密码", type="password", key=_ui_key("adm_u_pwd"))
+            if st.button("保存用户", type="primary", key=_ui_key("adm_u_save")):
+                if not u_name:
+                    st.error("用户名不能为空")
+                elif not u_pwd or len(u_pwd) < 6:
+                    st.error("密码至少 6 位")
+                else:
+                    uid = db.upsert_user(
+                        username=u_name.strip(),
+                        password=u_pwd,
+                        role=u_role,
+                        is_active=(1 if u_active == "启用" else 0),
+                    )
+                    st.success(f"已保存用户：{u_name} (id={uid})")
+                    st.rerun()
+
+        with st.expander("重置/禁用用户（按用户名）", expanded=False):
+            uname = st.selectbox(
+                "选择用户",
+                options=[u.get("username") for u in users if u.get("username")],
+                key=_ui_key("adm_sel_user"),
+            ) if users else None
+            new_pass = st.text_input("新密码（可选）", type="password", key=_ui_key("adm_reset_pwd"))
+            new_active = st.selectbox("状态", ["不修改", "启用", "禁用"], key=_ui_key("adm_reset_active"))
+            if st.button("应用修改", type="secondary", key=_ui_key("adm_apply")):
+                target = next((u for u in users if u.get("username") == uname), None)
+                if not target:
+                    st.error("未找到用户")
+                else:
+                    if new_pass:
+                        if len(new_pass) < 6:
+                            st.error("密码至少 6 位")
+                            st.stop()
+                        db.reset_user_password(target.get("id"), new_pass)
+                    if new_active != "不修改":
+                        db.set_user_active(target.get("id"), 1 if new_active == "启用" else 0)
+                    st.success("已更新")
             st.rerun()
 

assets/stickers_builtin/arrow.svg

@@ -0,0 +1,14 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="512" height="512" viewBox="0 0 512 512">
+  <defs>
+    <filter id="s" x="-20%" y="-20%" width="140%" height="140%">
+      <feDropShadow dx="0" dy="8" stdDeviation="10" flood-color="#000" flood-opacity="0.35"/>
+    </filter>
+  </defs>
+  <g filter="url(#s)" fill="none" stroke="#00E5FF" stroke-width="40" stroke-linecap="round" stroke-linejoin="round">
+    <path d="M120 320 C 220 240, 280 220, 392 168" />
+    <path d="M340 120 L 412 164 L 356 236" />
+  </g>
+</svg>
+
+
+

assets/stickers_builtin/arrow_curve.svg

@@ -0,0 +1,14 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="512" height="512" viewBox="0 0 512 512">
+  <defs>
+    <filter id="s" x="-20%" y="-20%" width="140%" height="140%">
+      <feDropShadow dx="0" dy="10" stdDeviation="12" flood-color="#000" flood-opacity="0.35"/>
+    </filter>
+  </defs>
+  <g filter="url(#s)" fill="none" stroke="#00E676" stroke-width="44" stroke-linecap="round" stroke-linejoin="round">
+    <path d="M130 360 C 160 220, 260 180, 360 150" />
+    <path d="M320 96 L402 142 L346 226" />
+  </g>
+</svg>
+
+
+

assets/stickers_builtin/arrow_down.svg

@@ -0,0 +1,13 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="512" height="512" viewBox="0 0 512 512">
+  <defs>
+    <filter id="s" x="-20%" y="-20%" width="140%" height="140%">
+      <feDropShadow dx="0" dy="12" stdDeviation="14" flood-color="#000" flood-opacity="0.35"/>
+    </filter>
+  </defs>
+  <g filter="url(#s)" fill="#FF2E88">
+    <path d="M256 442 L402 292 h-86 V70 H196 v222 h-86 Z"/>
+  </g>
+</svg>
+
+
+

assets/stickers_builtin/arrow_left.svg

@@ -0,0 +1,13 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="512" height="512" viewBox="0 0 512 512">
+  <defs>
+    <filter id="s" x="-20%" y="-20%" width="140%" height="140%">
+      <feDropShadow dx="0" dy="12" stdDeviation="14" flood-color="#000" flood-opacity="0.35"/>
+    </filter>
+  </defs>
+  <g filter="url(#s)" fill="#7C3AED">
+    <path d="M70 256 L220 110 v86 h222 v120 H220 v86 Z"/>
+  </g>
+</svg>
+
+
+

assets/stickers_builtin/arrow_right.svg

@@ -0,0 +1,13 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="512" height="512" viewBox="0 0 512 512">
+  <defs>
+    <filter id="s" x="-20%" y="-20%" width="140%" height="140%">
+      <feDropShadow dx="0" dy="12" stdDeviation="14" flood-color="#000" flood-opacity="0.35"/>
+    </filter>
+  </defs>
+  <g filter="url(#s)" fill="#FFD700">
+    <path d="M442 256 L292 402 v-86 H70 V196 h222 v-86 Z"/>
+  </g>
+</svg>
+
+
+

assets/stickers_builtin/arrow_up.svg

@@ -0,0 +1,13 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="512" height="512" viewBox="0 0 512 512">
+  <defs>
+    <filter id="s" x="-20%" y="-20%" width="140%" height="140%">
+      <feDropShadow dx="0" dy="12" stdDeviation="14" flood-color="#000" flood-opacity="0.35"/>
+    </filter>
+  </defs>
+  <g filter="url(#s)" fill="#00E5FF">
+    <path d="M256 70 L110 220 h86 v222 h120 V220 h86 Z"/>
+  </g>
+</svg>
+
+
+

assets/stickers_builtin/benefit.svg

@@ -0,0 +1,21 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="512" height="512" viewBox="0 0 512 512">
+  <defs>
+    <linearGradient id="g" x1="0" y1="0" x2="1" y2="1">
+      <stop offset="0" stop-color="#FF2E88"/>
+      <stop offset="1" stop-color="#7C3AED"/>
+    </linearGradient>
+    <filter id="s" x="-20%" y="-20%" width="140%" height="140%">
+      <feDropShadow dx="0" dy="12" stdDeviation="14" flood-color="#000" flood-opacity="0.35"/>
+    </filter>
+  </defs>
+  <g filter="url(#s)">
+    <rect x="92" y="150" width="328" height="212" rx="44" fill="url(#g)"/>
+    <path d="M140 190h232" stroke="#fff" stroke-width="18" stroke-linecap="round" opacity="0.55"/>
+    <text x="256" y="295" font-size="84" text-anchor="middle" font-family="Arial Black, Arial" fill="#fff">福利</text>
+    <circle cx="160" cy="330" r="20" fill="#FFD700"/>
+    <circle cx="352" cy="330" r="20" fill="#FFD700"/>
+  </g>
+</svg>
+
+
+

assets/stickers_builtin/bubble_buy.svg

@@ -0,0 +1,19 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="512" height="512" viewBox="0 0 512 512">
+  <defs>
+    <linearGradient id="g" x1="0" y1="0" x2="1" y2="1">
+      <stop offset="0" stop-color="#FFD700"/>
+      <stop offset="1" stop-color="#FF3D00"/>
+    </linearGradient>
+    <filter id="s" x="-25%" y="-25%" width="150%" height="150%">
+      <feDropShadow dx="0" dy="10" stdDeviation="12" flood-color="#000" flood-opacity="0.35"/>
+    </filter>
+  </defs>
+  <g filter="url(#s)">
+    <path d="M92 176c0-44 36-80 80-80h168c92 0 166 74 166 166s-74 166-166 166H258l-70 60 18-60h-34c-44 0-80-36-80-80V176z" fill="#111" opacity="0.18"/>
+    <path d="M92 160c0-44 36-80 80-80h168c92 0 166 74 166 166s-74 166-166 166H258l-70 60 18-60h-34c-44 0-80-36-80-80V160z" fill="url(#g)"/>
+    <text x="296" y="292" font-size="84" text-anchor="middle" font-family="Arial Black, Arial" fill="#111">买它</text>
+  </g>
+</svg>
+
+
+

assets/stickers_builtin/bubble_go.svg

@@ -0,0 +1,15 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="512" height="512" viewBox="0 0 512 512">
+  <defs>
+    <filter id="s" x="-25%" y="-25%" width="150%" height="150%">
+      <feDropShadow dx="0" dy="10" stdDeviation="12" flood-color="#000" flood-opacity="0.35"/>
+    </filter>
+  </defs>
+  <g filter="url(#s)">
+    <path d="M92 170c0-40 32-72 72-72h184c84 0 152 68 152 152s-68 152-152 152H260l-72 62 18-62h-42c-40 0-72-32-72-72V170z" fill="#111" opacity="0.2"/>
+    <path d="M92 156c0-40 32-72 72-72h184c84 0 152 68 152 152s-68 152-152 152H260l-72 62 18-62h-42c-40 0-72-32-72-72V156z" fill="#FF2E88"/>
+    <text x="296" y="286" font-size="92" text-anchor="middle" font-family="Arial Black, Arial" fill="#fff">冲！</text>
+  </g>
+</svg>
+
+
+

assets/stickers_builtin/bubble_link.svg

@@ -0,0 +1,19 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="512" height="512" viewBox="0 0 512 512">
+  <defs>
+    <linearGradient id="g" x1="0" y1="0" x2="1" y2="1">
+      <stop offset="0" stop-color="#00E5FF"/>
+      <stop offset="1" stop-color="#FF2E88"/>
+    </linearGradient>
+    <filter id="s" x="-25%" y="-25%" width="150%" height="150%">
+      <feDropShadow dx="0" dy="10" stdDeviation="12" flood-color="#000" flood-opacity="0.35"/>
+    </filter>
+  </defs>
+  <g filter="url(#s)">
+    <path d="M92 172c0-44 36-80 80-80h176c92 0 166 74 166 166s-74 166-166 166H260l-76 66 20-66h-32c-44 0-80-36-80-80V172z" fill="url(#g)"/>
+    <text x="300" y="292" font-size="78" text-anchor="middle" font-family="Arial Black, Arial" fill="#111">上链接</text>
+    <path d="M356 210c18-18 46-18 64 0s18 46 0 64l-22 22c-18 18-46 18-64 0" fill="none" stroke="#111" stroke-width="14" stroke-linecap="round"/>
+  </g>
+</svg>
+
+
+

assets/stickers_builtin/bubble_nice.svg

@@ -0,0 +1,15 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="512" height="512" viewBox="0 0 512 512">
+  <defs>
+    <filter id="s" x="-25%" y="-25%" width="150%" height="150%">
+      <feDropShadow dx="0" dy="10" stdDeviation="12" flood-color="#000" flood-opacity="0.35"/>
+    </filter>
+  </defs>
+  <g filter="url(#s)">
+    <path d="M96 176c0-44 36-80 80-80h168c92 0 166 74 166 166s-74 166-166 166H252l-78 66 20-66h-18c-44 0-80-36-80-80V176z" fill="#00E676"/>
+    <text x="292" y="292" font-size="84" text-anchor="middle" font-family="Arial Black, Arial" fill="#111">真香</text>
+    <path d="M166 126l26 22-20 30" fill="none" stroke="#111" stroke-width="12" stroke-linecap="round" stroke-linejoin="round" opacity="0.35"/>
+  </g>
+</svg>
+
+
+

assets/stickers_builtin/bubble_order.svg

@@ -0,0 +1,18 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="512" height="512" viewBox="0 0 512 512">
+  <defs>
+    <linearGradient id="g" x1="0" y1="0" x2="1" y2="1">
+      <stop offset="0" stop-color="#7C3AED"/>
+      <stop offset="1" stop-color="#00E5FF"/>
+    </linearGradient>
+    <filter id="s" x="-25%" y="-25%" width="150%" height="150%">
+      <feDropShadow dx="0" dy="10" stdDeviation="12" flood-color="#000" flood-opacity="0.35"/>
+    </filter>
+  </defs>
+  <g filter="url(#s)">
+    <path d="M86 170c0-44 36-80 80-80h180c92 0 166 74 166 166s-74 166-166 166H256l-84 72 22-72h-28c-44 0-80-36-80-80V170z" fill="url(#g)"/>
+    <text x="300" y="294" font-size="84" text-anchor="middle" font-family="Arial Black, Arial" fill="#111">安排</text>
+  </g>
+</svg>
+
+
+

assets/stickers_builtin/coupon.svg

@@ -0,0 +1,19 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="512" height="512" viewBox="0 0 512 512">
+  <defs>
+    <linearGradient id="g" x1="0" y1="0" x2="1" y2="1">
+      <stop offset="0" stop-color="#00E676"/>
+      <stop offset="1" stop-color="#00E5FF"/>
+    </linearGradient>
+    <filter id="s" x="-20%" y="-20%" width="140%" height="140%">
+      <feDropShadow dx="0" dy="12" stdDeviation="14" flood-color="#000" flood-opacity="0.35"/>
+    </filter>
+  </defs>
+  <g filter="url(#s)">
+    <path d="M110 170h292c20 0 36 16 36 36v28c-22 0-40 18-40 40s18 40 40 40v28c0 20-16 36-36 36H110c-20 0-36-16-36-36v-28c22 0 40-18 40-40s-18-40-40-40v-28c0-20 16-36 36-36z" fill="url(#g)"/>
+    <path d="M210 190v236" stroke="#fff" stroke-width="14" stroke-dasharray="14 14" opacity="0.7"/>
+    <text x="320" y="300" font-size="76" text-anchor="middle" font-family="Arial Black, Arial" fill="#111">领券</text>
+  </g>
+</svg>
+
+
+

assets/stickers_builtin/follow.svg

@@ -0,0 +1,21 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="512" height="512" viewBox="0 0 512 512">
+  <defs>
+    <linearGradient id="g" x1="0" y1="0" x2="1" y2="1">
+      <stop offset="0" stop-color="#00E5FF"/>
+      <stop offset="1" stop-color="#7C3AED"/>
+    </linearGradient>
+    <filter id="s" x="-20%" y="-20%" width="140%" height="140%">
+      <feDropShadow dx="0" dy="12" stdDeviation="14" flood-color="#000" flood-opacity="0.35"/>
+    </filter>
+  </defs>
+  <g filter="url(#s)">
+    <rect x="60" y="170" width="392" height="172" rx="40" fill="url(#g)"/>
+    <rect x="72" y="182" width="368" height="148" rx="34" fill="#111" opacity="0.18"/>
+    <text x="256" y="285" font-size="88" text-anchor="middle" font-family="Arial Black, Arial" fill="#fff">关注</text>
+    <circle cx="410" cy="256" r="42" fill="#FF2E88"/>
+    <path d="M410 234v44M388 256h44" stroke="#fff" stroke-width="16" stroke-linecap="round"/>
+  </g>
+</svg>
+
+
+

assets/stickers_builtin/hot.svg

@@ -0,0 +1,19 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="512" height="512" viewBox="0 0 512 512">
+  <defs>
+    <linearGradient id="g" x1="0" y1="0" x2="1" y2="1">
+      <stop offset="0" stop-color="#FF3D00"/>
+      <stop offset="1" stop-color="#FFD700"/>
+    </linearGradient>
+    <filter id="s" x="-25%" y="-25%" width="150%" height="150%">
+      <feDropShadow dx="0" dy="14" stdDeviation="16" flood-color="#000" flood-opacity="0.35"/>
+    </filter>
+  </defs>
+  <g filter="url(#s)">
+    <path d="M110 210 Q256 90 402 210 Q452 252 430 318 Q395 420 256 440 Q117 420 82 318 Q60 252 110 210 Z" fill="url(#g)"/>
+    <path d="M256 150c-18 26-8 44 10 64 18 20 28 38 18 62-10 26-36 32-56 18 6 34 42 58 82 40 38-18 48-62 28-96-18-32-42-48-38-88 2-18 10-28 18-38-28 8-46 22-62 38z" fill="#fff" opacity="0.9"/>
+    <text x="256" y="318" font-size="84" text-anchor="middle" font-family="Arial Black, Arial" fill="#111">爆款</text>
+  </g>
+</svg>
+
+
+

assets/stickers_builtin/index.json

@@ -0,0 +1,53 @@
+{
+  "pack": {
+    "id": "builtin-basic",
+    "name": "内置贴纸库（中国抖音常用）",
+    "license": "Project Built-in (owned)",
+    "attribution": "Built-in sticker pack generated by Video Flow. No external attribution required."
+  },
+  "categories": [
+    {
+      "id": "douyin-basic",
+      "name": "抖音常用",
+      "items": [
+        { "id": "like", "name": "点赞", "file": "like.svg", "tags": ["点赞", "爱心", "互动"] },
+        { "id": "follow", "name": "关注", "file": "follow.svg", "tags": ["关注", "订阅", "点赞关注"] },
+        { "id": "hot", "name": "爆款", "file": "hot.svg", "tags": ["爆款", "热卖", "火"] },
+        { "id": "new", "name": "新品", "file": "new.svg", "tags": ["新品", "上新"] },
+        { "id": "benefit", "name": "福利", "file": "benefit.svg", "tags": ["福利", "赠品", "优惠"] },
+        { "id": "sale", "name": "优惠", "file": "sale.svg", "tags": ["优惠", "促销", "价格"] },
+        { "id": "coupon", "name": "领券", "file": "coupon.svg", "tags": ["领券", "优惠券", "券"] },
+        { "id": "limit", "name": "限时", "file": "limit.svg", "tags": ["限时", "抢购", "倒计时"] },
+        { "id": "lowest", "name": "到手价", "file": "lowest.svg", "tags": ["到手价", "价格", "低价"] }
+      ]
+    }
+    ,
+    {
+      "id": "arrows",
+      "name": "箭头指引",
+      "items": [
+        { "id": "arrow-up", "name": "上箭头", "file": "arrow_up.svg", "tags": ["箭头", "指向", "上"] },
+        { "id": "arrow-down", "name": "下箭头", "file": "arrow_down.svg", "tags": ["箭头", "指向", "下"] },
+        { "id": "arrow-left", "name": "左箭头", "file": "arrow_left.svg", "tags": ["箭头", "指向", "左"] },
+        { "id": "arrow-right", "name": "右箭头", "file": "arrow_right.svg", "tags": ["箭头", "指向", "右"] },
+        { "id": "arrow-curve", "name": "弯箭头", "file": "arrow_curve.svg", "tags": ["箭头", "指向", "弯"] },
+        { "id": "pointer", "name": "手指", "file": "pointer.svg", "tags": ["手指", "指向", "点击"] },
+        { "id": "tap", "name": "戳这里", "file": "tap.svg", "tags": ["戳这里", "引导", "点击"] },
+        { "id": "look", "name": "看这里", "file": "look.svg", "tags": ["看这里", "引导", "注意"] }
+      ]
+    },
+    {
+      "id": "bubbles",
+      "name": "气泡弹幕",
+      "items": [
+        { "id": "bubble-1", "name": "气泡-冲！", "file": "bubble_go.svg", "tags": ["气泡", "弹幕", "冲"] },
+        { "id": "bubble-2", "name": "气泡-安排", "file": "bubble_order.svg", "tags": ["气泡", "弹幕", "安排"] },
+        { "id": "bubble-3", "name": "气泡-买它", "file": "bubble_buy.svg", "tags": ["气泡", "弹幕", "买它"] },
+        { "id": "bubble-4", "name": "气泡-真香", "file": "bubble_nice.svg", "tags": ["气泡", "弹幕", "真香"] },
+        { "id": "bubble-5", "name": "气泡-上链接", "file": "bubble_link.svg", "tags": ["气泡", "弹幕", "链接"] }
+      ]
+    }
+  ]
+}
+
+

assets/stickers_builtin/like.svg

@@ -0,0 +1,17 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="512" height="512" viewBox="0 0 512 512">
+  <defs>
+    <linearGradient id="g" x1="0" y1="0" x2="1" y2="1">
+      <stop offset="0" stop-color="#FF2E88"/>
+      <stop offset="1" stop-color="#FF7A00"/>
+    </linearGradient>
+    <filter id="s" x="-20%" y="-20%" width="140%" height="140%">
+      <feDropShadow dx="0" dy="10" stdDeviation="12" flood-color="#000" flood-opacity="0.35"/>
+    </filter>
+  </defs>
+  <g filter="url(#s)">
+    <path fill="url(#g)" d="M256 456s-148-88-196-176C18 208 36 132 106 104c48-19 100-6 138 34 38-40 90-53 138-34 70 28 88 104 46 176-48 88-196 176-196 176z"/>
+  </g>
+</svg>
+
+
+

assets/stickers_builtin/limit.svg

@@ -0,0 +1,22 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="512" height="512" viewBox="0 0 512 512">
+  <defs>
+    <linearGradient id="g" x1="0" y1="0" x2="1" y2="1">
+      <stop offset="0" stop-color="#FF3D00"/>
+      <stop offset="1" stop-color="#FF2E88"/>
+    </linearGradient>
+    <filter id="s" x="-20%" y="-20%" width="140%" height="140%">
+      <feDropShadow dx="0" dy="12" stdDeviation="14" flood-color="#000" flood-opacity="0.35"/>
+    </filter>
+  </defs>
+  <g filter="url(#s)">
+    <rect x="88" y="160" width="336" height="192" rx="44" fill="url(#g)"/>
+    <text x="256" y="270" font-size="84" text-anchor="middle" font-family="Arial Black, Arial" fill="#fff">限时</text>
+    <g transform="translate(140,310)">
+      <rect x="0" y="0" width="232" height="46" rx="18" fill="#111" opacity="0.25"/>
+      <text x="116" y="34" font-size="26" text-anchor="middle" font-family="Arial Black, Arial" fill="#fff">抢购中</text>
+    </g>
+  </g>
+</svg>
+
+
+

assets/stickers_builtin/look.svg

@@ -0,0 +1,19 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="512" height="512" viewBox="0 0 512 512">
+  <defs>
+    <linearGradient id="g" x1="0" y1="0" x2="1" y2="1">
+      <stop offset="0" stop-color="#00E5FF"/>
+      <stop offset="1" stop-color="#00E676"/>
+    </linearGradient>
+    <filter id="s" x="-25%" y="-25%" width="150%" height="150%">
+      <feDropShadow dx="0" dy="12" stdDeviation="14" flood-color="#000" flood-opacity="0.35"/>
+    </filter>
+  </defs>
+  <g filter="url(#s)">
+    <rect x="88" y="196" width="336" height="160" rx="56" fill="url(#g)"/>
+    <text x="256" y="302" font-size="84" text-anchor="middle" font-family="Arial Black, Arial" fill="#111">看这里</text>
+    <path d="M120 170l44 28-20 52" fill="none" stroke="#111" stroke-width="14" stroke-linecap="round" stroke-linejoin="round" opacity="0.35"/>
+  </g>
+</svg>
+
+
+

assets/stickers_builtin/lowest.svg

@@ -0,0 +1,19 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="512" height="512" viewBox="0 0 512 512">
+  <defs>
+    <linearGradient id="g" x1="0" y1="0" x2="1" y2="1">
+      <stop offset="0" stop-color="#FFD700"/>
+      <stop offset="1" stop-color="#FF3D00"/>
+    </linearGradient>
+    <filter id="s" x="-25%" y="-25%" width="150%" height="150%">
+      <feDropShadow dx="0" dy="12" stdDeviation="14" flood-color="#000" flood-opacity="0.35"/>
+    </filter>
+  </defs>
+  <g filter="url(#s)">
+    <path d="M110 210 Q256 120 402 210 Q440 235 430 278 Q410 360 256 392 Q102 360 82 278 Q72 235 110 210 Z" fill="url(#g)"/>
+    <text x="256" y="300" font-size="72" text-anchor="middle" font-family="Arial Black, Arial" fill="#111">到手价</text>
+    <path d="M166 330h180" stroke="#111" stroke-width="16" stroke-linecap="round" opacity="0.35"/>
+  </g>
+</svg>
+
+
+

assets/stickers_builtin/new.svg

@@ -0,0 +1,18 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="512" height="512" viewBox="0 0 512 512">
+  <defs>
+    <linearGradient id="g" x1="0" y1="0" x2="1" y2="1">
+      <stop offset="0" stop-color="#7C3AED"/>
+      <stop offset="1" stop-color="#00E5FF"/>
+    </linearGradient>
+    <filter id="s" x="-20%" y="-20%" width="140%" height="140%">
+      <feDropShadow dx="0" dy="12" stdDeviation="14" flood-color="#000" flood-opacity="0.35"/>
+    </filter>
+  </defs>
+  <g filter="url(#s)">
+    <path d="M120 150h272l40 60-40 60H120l-40-60z" fill="url(#g)"/>
+    <text x="256" y="238" font-size="84" text-anchor="middle" font-family="Arial Black, Arial" fill="#fff">新品</text>
+  </g>
+</svg>
+
+
+

assets/stickers_builtin/pointer.svg

@@ -0,0 +1,16 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="512" height="512" viewBox="0 0 512 512">
+  <defs>
+    <filter id="s" x="-20%" y="-20%" width="140%" height="140%">
+      <feDropShadow dx="0" dy="12" stdDeviation="14" flood-color="#000" flood-opacity="0.35"/>
+    </filter>
+  </defs>
+  <g filter="url(#s)">
+    <path d="M220 72c22 0 40 18 40 40v152h22V150c0-20 16-36 36-36s36 16 36 36v114h20V182c0-18 14-32 32-32s32 14 32 32v160c0 52-42 94-94 94H260c-66 0-120-54-120-120v-8c0-22 18-40 40-40h40V112c0-22 18-40 40-40z" fill="#fff"/>
+    <path d="M220 72c22 0 40 18 40 40v152h22V150c0-20 16-36 36-36s36 16 36 36v114h20V182c0-18 14-32 32-32s32 14 32 32v160c0 52-42 94-94 94H260c-66 0-120-54-120-120v-8c0-22 18-40 40-40h40V112c0-22 18-40 40-40z" fill="none" stroke="#111" stroke-width="18" stroke-linejoin="round"/>
+    <circle cx="386" cy="92" r="32" fill="#FF2E88"/>
+    <path d="M386 76v32M370 92h32" stroke="#fff" stroke-width="12" stroke-linecap="round"/>
+  </g>
+</svg>
+
+
+

assets/stickers_builtin/sale.svg

@@ -0,0 +1,22 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="512" height="512" viewBox="0 0 512 512">
+  <defs>
+    <linearGradient id="g" x1="0" y1="0" x2="1" y2="1">
+      <stop offset="0" stop-color="#FFD700"/>
+      <stop offset="1" stop-color="#FF3D00"/>
+    </linearGradient>
+    <filter id="s" x="-20%" y="-20%" width="140%" height="140%">
+      <feDropShadow dx="0" dy="10" stdDeviation="12" flood-color="#000" flood-opacity="0.35"/>
+    </filter>
+  </defs>
+  <g filter="url(#s)">
+    <path fill="url(#g)" d="M110 160 L300 70 L430 200 L240 290 Z"/>
+    <circle cx="170" cy="200" r="18" fill="#fff"/>
+    <circle cx="310" cy="140" r="18" fill="#fff"/>
+    <path d="M210 222 L320 112" stroke="#fff" stroke-width="18" stroke-linecap="round"/>
+    <rect x="120" y="292" width="320" height="120" rx="28" fill="#111"/>
+    <text x="280" y="372" font-size="68" text-anchor="middle" font-family="Arial Black, Arial" fill="#fff">SALE</text>
+  </g>
+</svg>
+
+
+

assets/stickers_builtin/tap.svg

@@ -0,0 +1,18 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="512" height="512" viewBox="0 0 512 512">
+  <defs>
+    <linearGradient id="g" x1="0" y1="0" x2="1" y2="1">
+      <stop offset="0" stop-color="#FF2E88"/>
+      <stop offset="1" stop-color="#FFD700"/>
+    </linearGradient>
+    <filter id="s" x="-25%" y="-25%" width="150%" height="150%">
+      <feDropShadow dx="0" dy="12" stdDeviation="14" flood-color="#000" flood-opacity="0.35"/>
+    </filter>
+  </defs>
+  <g filter="url(#s)">
+    <path d="M88 210h336l-48 86 48 86H88l48-86z" fill="url(#g)"/>
+    <text x="256" y="314" font-size="78" text-anchor="middle" font-family="Arial Black, Arial" fill="#111">戳这里</text>
+  </g>
+</svg>
+
+
+

assets/stickers_builtin/wow.svg

@@ -0,0 +1,16 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="512" height="512" viewBox="0 0 512 512">
+  <defs>
+    <filter id="s" x="-20%" y="-20%" width="140%" height="140%">
+      <feDropShadow dx="0" dy="10" stdDeviation="12" flood-color="#000" flood-opacity="0.35"/>
+    </filter>
+  </defs>
+  <g filter="url(#s)">
+    <path d="M90 140 Q256 40 422 140 Q470 170 450 230 Q420 320 256 360 Q92 320 62 230 Q42 170 90 140 Z" fill="#7C3AED"/>
+    <path d="M130 170 Q256 90 382 170" fill="none" stroke="#fff" stroke-width="14" stroke-linecap="round" opacity="0.35"/>
+    <text x="256" y="265" font-size="96" text-anchor="middle" font-family="Arial Black, Arial" fill="#fff">WOW</text>
+    <path d="M200 320 L172 380 L240 350 Z" fill="#7C3AED"/>
+  </g>
+</svg>
+
+
+

config.py

@@ -38,6 +38,10 @@ DOUBAO_IMG_MODEL = "ep-20251203231641-wg9nb"
 SHUBIAOBIAO_KEY = os.getenv("SHUBIAOBIAO_KEY", "sk-aL167A8sQEyvs40yBfC140Fc0fDa4c198f029aAcF0429108")
 SHUBIAOBIAO_BASE_URL = os.getenv("SHUBIAOBIAO_BASE_URL", "https://api.shubiaobiao.cn/v1")
 SHUBIAOBIAO_MODEL_TEXT = "gemini-3-pro-preview"
+# ShuBiaoBiao (OpenAI-compatible) chat timeout/retries
+# Note: OpenAI Python SDK default timeout is 10 minutes; we cap it to avoid UI "hang" feeling.
+SHUBIAOBIAO_CHAT_TIMEOUT_S = float(os.getenv("SHUBIAOBIAO_CHAT_TIMEOUT_S", "180"))
+SHUBIAOBIAO_CHAT_MAX_RETRIES = int(os.getenv("SHUBIAOBIAO_CHAT_MAX_RETRIES", "2"))
 
 # Image Generation API (Updated)
 # Host: https://api.wuyinkeji.com/

docs/04-development/DEV-LEGACY-SCHEMA-20251215.md

@@ -0,0 +1,52 @@
+# Legacy Project JSON Schema Scan Report
+
+- temp_dir: `/opt/gloda-factory/temp`
+- total_files: 18
+- parsed_files: 18
+- failed_files: 0
+
+## Schema variants
+
+- Schema_A: 13 (samples: 0b0f819a, 18470131, 26ed8fa0, 3ce8a4ee, 61e70d91)
+- Unknown: 4 (samples: 01897830, 61a1e46d, 70663b6c, bf58ccd5)
+- Schema_B: 1 (samples: 690b2c54)
+
+## CTA type distribution
+
+- str: 17
+- dict: 1
+
+## Top-level keys (top 30)
+
+- id: 18/18
+- created_at: 18/18
+- status: 18/18
+- input_mode: 18/18
+- prompt: 18/18
+- image_urls: 18/18
+- video_url: 18/18
+- asr_text: 18/18
+- analysis: 18/18
+- questions: 18/18
+- answers: 18/18
+- hook: 18/18
+- scenes: 18/18
+- cta: 18/18
+- final_video_url: 18/18
+- bgm_url: 18/18
+
+## Scene keys (top 40)
+
+- id: 74
+- duration: 74
+- timeline: 74
+- camera_movement: 74
+- story_beat: 74
+- voiceover: 74
+- rhythm: 74
+- image_prompt: 69
+- keyframe: 69
+- sound_design: 69
+- image_url: 47
+- keyframes: 5
+

docs/07-user/USER-002-贴纸库数据源与授权.md

@@ -0,0 +1,51 @@
+## 目标
+为编辑器内置一个**适合抖音场景**的贴纸库（PNG/SVG），并保证：
+- **可商用/可分发**（许可清晰）
+- **可本地托管**（不依赖外部 CDN）
+- **所见即所得**：预览与导出一致（贴纸叠加到成片）
+
+## 推荐贴纸库（抖音场景友好）
+### 方案 A：Microsoft Fluent UI Emoji（更“抖音感”）
+- **风格**：高饱和、现代、偏 3D/大图标，适合“强调/氛围/卖点”
+- **形态**：PNG/SVG（仓库提供多种风格/尺寸）
+- **适用**：火/赞/心/星星/箭头/提示/表情等常用贴纸
+- **风险**：请在引入前再次核对仓库 LICENSE（不同仓库/分支可能不同）
+
+### 方案 B：Twemoji（稳定、覆盖全、但更像“emoji”）
+- **风格**：标准 emoji
+- **形态**：PNG/SVG
+- **适用**：作为“基础补全库”非常合适
+- **风险**：通常需要署名（CC-BY 类）；引入前核对 LICENSE
+
+### 不推荐（默认）
+### OpenMoji
+- **优点**：开源清晰、SVG 质量高
+- **缺点**：常见为 CC BY-SA（“同协议分享”约束强），对商业产品和二次分发不友好
+
+## 贴纸库落地方式（本项目）
+本项目支持两类贴纸：
+- **内置贴纸**：放在 `assets/stickers_builtin/`，通过 `assets/stickers_builtin/index.json` 声明分类/标签/授权信息。
+- **自定义贴纸**：用户上传到 `assets/stickers_custom/`，可直接在 UI 里使用。
+
+后端接口：
+- `GET /api/assets/stickers`：返回贴纸列表（合并 builtin + custom）
+- `POST /api/assets/stickers/upload`：上传 PNG/SVG/WEBP
+
+前端能力：
+- 左侧 **“贴纸”** Tab：搜索/分类/缩略图；**拖拽到时间轴**生成贴纸片段
+- 时间轴新增 **“贴纸”轨道**：贴纸片段可移动/裁剪时长
+- 右侧属性：贴纸 **大小/旋转/X/Y**
+
+导出（WYSIWYG）：
+- FFmpeg 叠加贴纸：`overlay_multiple_images`
+- SVG 会在导出侧被转换为 PNG（优先使用 `rsvg-convert`，Dockerfile 已加入 `librsvg2-bin`）
+
+## 下一步：把“推荐贴纸库”真正导入到 assets（需要一次性下载）
+由于贴纸库体积很大（数千~上万文件），建议用脚本把需要的子集同步到 `assets/stickers_builtin/`：
+- 先挑“抖音高频类目”：点赞/关注/箭头/爆款/促销/emoji 表情/弹幕气泡
+- 再逐步扩展
+
+我建议你确认最终选用的库（Fluent vs Twemoji）后，我可以给你一个“按清单下载 + 生成 index.json”的脚本（可在服务器执行）。
+
+
+

docs/EDITOR_README.md

@@ -0,0 +1,164 @@
+# Video Flow Editor - 视频编辑器
+
+## 概述
+
+Video Flow Editor 是一个基于浏览器的视频编辑器，支持多轨道时间轴编辑，与现有的 AI 视频生成工作流无缝集成。
+
+## 架构
+
+```
+┌─────────────────────────────────────────────────────────────┐
+│                    React Frontend (:3000)                    │
+│  ┌─────────────┐  ┌─────────────┐  ┌─────────────────────┐  │
+│  │  Remotion   │  │  Timeline   │  │    Track Panel      │  │
+│  │  预览播放器  │  │   时间轴    │  │   轨道属性面板      │  │
+│  └─────────────┘  └─────────────┘  └─────────────────────┘  │
+└─────────────────────────────────────────────────────────────┘
+                              │
+                              ▼
+┌─────────────────────────────────────────────────────────────┐
+│                   FastAPI Backend (:8000)                    │
+│  ┌─────────────┐  ┌─────────────┐  ┌─────────────────────┐  │
+│  │  项目管理   │  │  编辑器 API  │  │     合成 API        │  │
+│  └─────────────┘  └─────────────┘  └─────────────────────┘  │
+└─────────────────────────────────────────────────────────────┘
+                              │
+                              ▼
+┌─────────────────────────────────────────────────────────────┐
+│                    Celery + Redis                            │
+│  ┌─────────────┐  ┌─────────────┐  ┌─────────────────────┐  │
+│  │  任务队列   │  │  Worker 1   │  │    Worker N...      │  │
+│  └─────────────┘  └─────────────┘  └─────────────────────┘  │
+└─────────────────────────────────────────────────────────────┘
+```
+
+## 功能特性
+
+### 编辑功能
+- ✅ 视频裁剪 - 在时间轴上拖拽调整视频起止点
+- ✅ 旁白编辑 - 修改文本后重新生成 TTS
+- ✅ 花字编辑 - 修改花字内容和样式
+- ✅ BGM 管理 - 选择和替换背景音乐
+- ✅ 字幕编辑 - 修改字幕文本和时间
+
+### 预览功能
+- ✅ Remotion 实时预览 - 浏览器端实时渲染
+- ✅ 多轨道显示 - 视频/音频/字幕/花字/BGM
+- ✅ 播放头控制 - 拖拽快速定位
+
+### 导出功能
+- ✅ 异步合成 - 任务队列处理，不阻塞界面
+- ✅ 进度查询 - 实时查看合成进度
+- ✅ 下载成片 - 合成完成后直接下载
+
+## 快速开始
+
+### 开发环境
+
+```bash
+# 1. 安装依赖
+pip install -r requirements.txt
+cd web && npm install && cd ..
+
+# 2. 启动 Redis (需要 Docker)
+docker run -d --name redis -p 6379:6379 redis:7-alpine
+
+# 3. 启动后端
+uvicorn api.main:app --reload --port 8000
+
+# 4. 启动 Worker
+celery -A api.celery_app worker --loglevel=info
+
+# 5. 启动前端
+cd web && npm run dev
+```
+
+### Docker 环境
+
+```bash
+# 一键启动所有服务
+docker-compose up -d
+
+# 扩展 Worker 数量
+docker-compose scale worker=3
+
+# 查看日志
+docker-compose logs -f worker
+```
+
+## 端口规划
+
+| 服务 | 端口 | 说明 |
+|------|------|------|
+| React Editor | 3000 | 视频编辑器前端 |
+| FastAPI | 8000 | REST API |
+| Streamlit | 8502 | 原有工作流调试界面 |
+| Redis | 6379 | 任务队列 |
+
+## API 接口
+
+### 编辑器状态
+
+```
+GET  /api/editor/{project_id}/state  - 获取编辑器状态
+POST /api/editor/{project_id}/state  - 保存编辑器状态
+```
+
+### TTS 生成
+
+```
+POST /api/editor/generate-voiceover
+{
+    "text": "要转换的文本",
+    "voice_type": "zh_female_santongyongns_saturn_bigtts",
+    "target_duration": 3.0  // 可选
+}
+```
+
+### 视频合成
+
+```
+POST /api/compose/render
+{
+    "project_id": "PROJ-xxx",
+    "video_clips": [...],
+    "voiceover_clips": [...],
+    "subtitle_clips": [...],
+    "fancy_text_clips": [...],
+    "bgm_clip": {...}
+}
+
+GET /api/compose/status/{task_id}  - 查询合成进度
+```
+
+## 扩展性
+
+### 水平扩展
+
+```bash
+# 增加 Worker 数量
+docker-compose scale worker=5
+```
+
+### 性能监控
+
+- Celery Flower: `celery -A api.celery_app flower`
+- Redis 监控: `redis-cli monitor`
+
+## 与工作流的集成
+
+编辑器自动读取工作流生成的素材：
+
+1. 用户在 Streamlit 完成视频生成工作流
+2. 点击"编辑"按钮跳转到编辑器
+3. 编辑器自动加载项目的所有素材到时间轴
+4. 用户进行精细编辑
+5. 导出最终成品
+
+## 技术栈
+
+- **前端**: React 18 + TypeScript + Vite + Remotion
+- **后端**: FastAPI + Celery + Redis
+- **视频处理**: FFmpeg (Worker 中运行)
+- **数据库**: SQLite / PostgreSQL
+

modules/auth.py

@@ -0,0 +1,48 @@
+"""
+Auth helpers: password hashing + cookie token hashing.
+
+We intentionally avoid heavy dependencies. Password hashing uses PBKDF2-HMAC-SHA256.
+Session tokens are random and stored server-side as SHA256(token) hashes.
+"""
+
+from __future__ import annotations
+
+import hashlib
+import secrets
+from typing import Optional, Tuple
+
+
+PBKDF2_ITERS = 200_000
+
+
+def hash_password(password: str, salt_hex: Optional[str] = None) -> Tuple[str, str]:
+    salt = bytes.fromhex(salt_hex) if salt_hex else secrets.token_bytes(16)
+    dk = hashlib.pbkdf2_hmac("sha256", (password or "").encode("utf-8"), salt, PBKDF2_ITERS)
+    return dk.hex(), salt.hex()
+
+
+def verify_password(password: str, password_hash: str, salt_hex: str) -> bool:
+    cand, _ = hash_password(password, salt_hex=salt_hex)
+    return cand == (password_hash or "")
+
+
+def new_session_token() -> str:
+    return secrets.token_urlsafe(32)
+
+
+def hash_token(token: str) -> str:
+    return hashlib.sha256((token or "").encode("utf-8")).hexdigest()
+
+
+
+
+
+
+
+
+
+
+
+
+
+

modules/composer.py

@@ -502,11 +502,20 @@ class VideoComposer:
             current_video = fancy_path
             
         # 7. 添加 BGM
+        # 说明：add_bgm 的 ducking=True 路径使用 sidechaincompress，但该滤镜本身不做“混音”，
+        # 在某些 ffmpeg 版本/参数组合下会导致 BGM 听起来像“没加上”。
+        # 我们在 compose() 里已禁用 ducking，这里保持一致，使用 amix 叠加并提高默认音量。
         if bgm_path:
             bgm_output = str(Path(temp_root) / f"{output_name}_bgm.mp4")
             ffmpeg_utils.add_bgm(
-                current_video, bgm_path, bgm_output,
+                current_video,
-                bgm_volume=0.15
+                bgm_path,
+                bgm_output,
+                bgm_volume=0.20,
+                ducking=False,
+                duck_gain_db=-6.0,
+                fade_in=1.0,
+                fade_out=1.0,
             )
             self._add_temp(bgm_output)
             current_video = bgm_output

modules/db_manager.py

@@ -6,14 +6,43 @@
 import json
 import logging
 import time
-from typing import Dict, List, Any, Optional
+import secrets
+from typing import Dict, List, Any, Optional, Tuple
 
-from sqlalchemy import create_engine, Column, String, Integer, Text, Float, UniqueConstraint, func
+from sqlalchemy import create_engine, Column, String, Integer, Text, Float, UniqueConstraint, func, text, inspect
 from sqlalchemy.orm import sessionmaker, scoped_session, declarative_base
 from sqlalchemy.dialects.postgresql import JSONB
 
 import config
 
+# NOTE: Some deployments do not ship `modules/auth.py`.
+# Keep a minimal, local auth helper here to avoid hard dependency.
+import hashlib
+import hmac
+
+
+def _hash_password(password: str, salt_hex: str = None) -> tuple[str, str]:
+    salt = bytes.fromhex(salt_hex) if salt_hex else secrets.token_bytes(16)
+    dk = hashlib.pbkdf2_hmac("sha256", (password or "").encode("utf-8"), salt, 120_000)
+    return dk.hex(), salt.hex()
+
+
+def _verify_password(password: str, pwd_hash_hex: str, salt_hex: str) -> bool:
+    try:
+        cand, _ = _hash_password(password or "", salt_hex=salt_hex)
+        return hmac.compare_digest(cand, pwd_hash_hex or "")
+    except Exception:
+        return False
+
+
+def _new_session_token() -> str:
+    return secrets.token_urlsafe(32)
+
+
+def _hash_token(token: str) -> str:
+    # store only hash for sessions
+    return hashlib.sha256((token or "").encode("utf-8")).hexdigest()
+
 logger = logging.getLogger(__name__)
 
 Base = declarative_base()
@@ -26,6 +55,7 @@ class Project(Base):
     status = Column(String) # created, script_generated, images_generated, videos_generated, completed
     product_info = Column(Text) # JSON string (SQLite) or JSONB (PG - using Text for compat)
     script_data = Column(Text) # JSON string
+    owner_user_id = Column(String, index=True, nullable=True)
     created_at = Column(Float, default=time.time)
     updated_at = Column(Float, default=time.time, onupdate=time.time)
 
@@ -54,6 +84,62 @@ class AppConfig(Base):
     description = Column(Text, nullable=True)
     updated_at = Column(Float, default=time.time, onupdate=time.time)
 
+
+class User(Base):
+    __tablename__ = "users"
+    id = Column(String, primary_key=True)  # uuid hex
+    username = Column(String, unique=True, index=True)
+    password_hash = Column(String)
+    password_salt = Column(String)
+    role = Column(String, default="user")  # admin/user
+    is_active = Column(Integer, default=1)  # 1/0 for portability
+    created_at = Column(Float, default=time.time)
+    updated_at = Column(Float, default=time.time, onupdate=time.time)
+    last_login_at = Column(Float, nullable=True)
+
+
+class UserSession(Base):
+    __tablename__ = "user_sessions"
+    id = Column(String, primary_key=True)  # uuid hex
+    user_id = Column(String, index=True)
+    token_hash = Column(String, index=True)
+    expires_at = Column(Float)
+    created_at = Column(Float, default=time.time)
+    last_seen_at = Column(Float, default=time.time)
+    ip = Column(String, nullable=True)
+    user_agent = Column(String, nullable=True)
+
+
+class UserPrompt(Base):
+    __tablename__ = "user_prompts"
+    id = Column(String, primary_key=True)  # uuid hex
+    user_id = Column(String, index=True)
+    key = Column(String, index=True)
+    value = Column(Text)
+    updated_at = Column(Float, default=time.time, onupdate=time.time)
+    __table_args__ = (UniqueConstraint("user_id", "key", name="uix_user_prompt"),)
+
+
+class RenderJob(Base):
+    """
+    Render job status table (async compose pipeline).
+    This is intentionally minimal and portable across SQLite/Postgres.
+    """
+    __tablename__ = "render_jobs"
+
+    id = Column(String, primary_key=True)  # task_id
+    project_id = Column(String, index=True)
+    status = Column(String, default="queued")  # queued/running/success/failed/cancelled
+    progress = Column(Float, default=0.0)
+    message = Column(Text, default="")
+    output_path = Column(Text, nullable=True)
+    output_url = Column(Text, nullable=True)
+    error = Column(Text, nullable=True)
+    request_json = Column(Text, nullable=True)  # JSON string of ComposeRequest
+    parent_id = Column(String, nullable=True, index=True)
+    created_at = Column(Float, default=time.time)
+    updated_at = Column(Float, default=time.time, onupdate=time.time)
+
 class DBManager:
     def __init__(self, connection_string: str = None):
         if not connection_string:
@@ -62,17 +148,59 @@ class DBManager:
         self.engine = create_engine(connection_string, pool_recycle=3600)
         self.Session = scoped_session(sessionmaker(bind=self.engine))
         self._init_db()
+        # bootstrap admin (safe to call repeatedly)
+        try:
+            self._bootstrap_admin_id = self.ensure_admin_user("admin", "admin1234")
+        except Exception:
+            self._bootstrap_admin_id = None
 
     def _init_db(self):
-        """初始化表结构"""
+        """初始化表结构 + 轻量自迁移（不依赖 Alembic）"""
         Base.metadata.create_all(self.engine)
+        self._ensure_schema()
+
+    def _ensure_schema(self) -> None:
+        """
+        Ensure newer columns/tables exist in both Postgres and SQLite.
+        create_all will not add columns to existing tables, so we do a minimal ALTER here.
+        """
+        try:
+            insp = inspect(self.engine)
+            # projects.owner_user_id
+            try:
+                cols = [c["name"] for c in insp.get_columns("projects")]
+            except Exception:
+                cols = []
+            if "owner_user_id" not in cols:
+                logger.info("Migrating: add projects.owner_user_id")
+                with self.engine.begin() as conn:
+                    conn.execute(text("ALTER TABLE projects ADD COLUMN owner_user_id VARCHAR"))
+                    try:
+                        conn.execute(text("CREATE INDEX IF NOT EXISTS ix_projects_owner_user_id ON projects (owner_user_id)"))
+                    except Exception:
+                        # some engines (older sqlite) may not support IF NOT EXISTS
+                        try:
+                            conn.execute(text("CREATE INDEX ix_projects_owner_user_id ON projects (owner_user_id)"))
+                        except Exception:
+                            pass
+        except Exception as e:
+            logger.warning(f"Schema ensure skipped/failed (non-fatal): {e}")
+
+        # Ensure render_jobs table exists (create_all should handle this; keep for safety)
+        try:
+            insp = inspect(self.engine)
+            if "render_jobs" not in insp.get_table_names():
+                logger.info("Migrating: create render_jobs table")
+                Base.metadata.create_all(self.engine)
+        except Exception as e:
+            logger.warning(f"render_jobs ensure skipped/failed (non-fatal): {e}")
 
     def _get_session(self):
         return self.Session()
 
     # --- Project Operations ---
 
-    def create_project(self, project_id: str, name: str, product_info: Dict[str, Any]):
+    def create_project(self, project_id: str, name: str, product_info: Dict[str, Any], owner_user_id: Optional[str] = None):
         session = self._get_session()
         try:
             # Check if exists
@@ -86,6 +214,7 @@ class DBManager:
                 name=name,
                 status="created",
                 product_info=json.dumps(product_info, ensure_ascii=False),
+                owner_user_id=owner_user_id,
                 created_at=time.time(),
                 updated_at=time.time()
             )
@@ -157,6 +286,7 @@ class DBManager:
                     "status": project.status,
                     "product_info": json.loads(project.product_info) if project.product_info else {},
                     "script_data": json.loads(project.script_data) if project.script_data else None,
+                    "owner_user_id": getattr(project, "owner_user_id", None),
                     "created_at": project.created_at,
                     "updated_at": project.updated_at
                 }
@@ -175,12 +305,288 @@ class DBManager:
                     "id": p.id,
                     "name": p.name,
                     "status": p.status,
+                    "owner_user_id": getattr(p, "owner_user_id", None),
                     "updated_at": p.updated_at
                 })
             return results
         finally:
             session.close()
 
+    # --- User/Auth Operations ---
+
+    def ensure_admin_user(self, username: str = "admin", password: str = "admin1234") -> str:
+        """Create bootstrap admin user if missing. Returns admin user_id."""
+        session = self._get_session()
+        try:
+            u = session.query(User).filter_by(username=username).first()
+            if u:
+                return u.id
+            pwd_hash, salt_hex = _hash_password(password)
+            uid = secrets.token_hex(16)
+            u = User(
+                id=uid,
+                username=username,
+                password_hash=pwd_hash,
+                password_salt=salt_hex,
+                role="admin",
+                is_active=1,
+                created_at=time.time(),
+                updated_at=time.time(),
+            )
+            session.add(u)
+            session.commit()
+            logger.warning("Bootstrap admin created (please change password asap).")
+            return uid
+        except Exception as e:
+            session.rollback()
+            logger.error(f"ensure_admin_user failed: {e}")
+            raise
+        finally:
+            session.close()
+
+    def authenticate_user(self, username: str, password: str) -> Optional[Dict[str, Any]]:
+        session = self._get_session()
+        try:
+            u = session.query(User).filter_by(username=username).first()
+            if not u or int(getattr(u, "is_active", 1) or 0) != 1:
+                return None
+            if not _verify_password(password or "", getattr(u, "password_hash", ""), getattr(u, "password_salt", "")):
+                return None
+            u.last_login_at = time.time()
+            session.commit()
+            return {"id": u.id, "username": u.username, "role": u.role, "is_active": int(u.is_active or 0)}
+        except Exception as e:
+            session.rollback()
+            logger.error(f"authenticate_user error: {e}")
+            return None
+        finally:
+            session.close()
+
+    def create_session(self, user_id: str, *, ttl_seconds: int = 7 * 24 * 3600, ip: str = None, user_agent: str = None) -> str:
+        """Returns raw session token (store hash in DB)."""
+        token = _new_session_token()
+        token_hash = _hash_token(token)
+        sid = secrets.token_hex(16)
+        session = self._get_session()
+        try:
+            now = time.time()
+            s = UserSession(
+                id=sid,
+                user_id=user_id,
+                token_hash=token_hash,
+                expires_at=now + int(ttl_seconds),
+                created_at=now,
+                last_seen_at=now,
+                ip=ip,
+                user_agent=user_agent,
+            )
+            session.add(s)
+            session.commit()
+            return token
+        except Exception as e:
+            session.rollback()
+            logger.error(f"create_session error: {e}")
+            raise
+        finally:
+            session.close()
+
+    def validate_session(self, token: str) -> Optional[Dict[str, Any]]:
+        if not token:
+            return None
+        token_hash = _hash_token(token)
+        session = self._get_session()
+        try:
+            now = time.time()
+            s = session.query(UserSession).filter_by(token_hash=token_hash).first()
+            if not s or (s.expires_at and float(s.expires_at) < now):
+                return None
+            u = session.query(User).filter_by(id=s.user_id).first()
+            if not u or int(getattr(u, "is_active", 1) or 0) != 1:
+                return None
+            s.last_seen_at = now
+            session.commit()
+            return {"id": u.id, "username": u.username, "role": u.role, "is_active": int(u.is_active or 0)}
+        except Exception as e:
+            session.rollback()
+            logger.error(f"validate_session error: {e}")
+            return None
+        finally:
+            session.close()
+
+    def revoke_session(self, token: str) -> None:
+        if not token:
+            return
+        token_hash = _hash_token(token)
+        session = self._get_session()
+        try:
+            s = session.query(UserSession).filter_by(token_hash=token_hash).first()
+            if s:
+                session.delete(s)
+                session.commit()
+        except Exception as e:
+            session.rollback()
+            logger.error(f"revoke_session error: {e}")
+        finally:
+            session.close()
+
+    def list_users(self) -> List[Dict[str, Any]]:
+        session = self._get_session()
+        try:
+            users = session.query(User).order_by(User.created_at.asc()).all()
+            return [
+                {
+                    "id": u.id,
+                    "username": u.username,
+                    "role": u.role,
+                    "is_active": int(u.is_active or 0),
+                    "created_at": u.created_at,
+                    "last_login_at": u.last_login_at,
+                }
+                for u in users
+            ]
+        finally:
+            session.close()
+
+    def upsert_user(self, username: str, password: str, *, role: str = "user", is_active: int = 1) -> str:
+        session = self._get_session()
+        try:
+            u = session.query(User).filter_by(username=username).first()
+            pwd_hash, salt_hex = _hash_password(password)
+            now = time.time()
+            if u:
+                u.password_hash = pwd_hash
+                u.password_salt = salt_hex
+                u.role = role
+                u.is_active = int(is_active)
+                u.updated_at = now
+                session.commit()
+                return u.id
+            uid = secrets.token_hex(16)
+            u = User(
+                id=uid,
+                username=username,
+                password_hash=pwd_hash,
+                password_salt=salt_hex,
+                role=role,
+                is_active=int(is_active),
+                created_at=now,
+                updated_at=now,
+            )
+            session.add(u)
+            session.commit()
+            return uid
+        except Exception as e:
+            session.rollback()
+            logger.error(f"upsert_user error: {e}")
+            raise
+        finally:
+            session.close()
+
+    def set_user_active(self, user_id: str, is_active: int) -> None:
+        session = self._get_session()
+        try:
+            u = session.query(User).filter_by(id=user_id).first()
+            if not u:
+                return
+            u.is_active = int(is_active)
+            u.updated_at = time.time()
+            session.commit()
+        except Exception as e:
+            session.rollback()
+            logger.error(f"set_user_active error: {e}")
+        finally:
+            session.close()
+
+    def reset_user_password(self, user_id: str, new_password: str) -> None:
+        session = self._get_session()
+        try:
+            u = session.query(User).filter_by(id=user_id).first()
+            if not u:
+                return
+            pwd_hash, salt_hex = _hash_password(new_password)
+            u.password_hash = pwd_hash
+            u.password_salt = salt_hex
+            u.updated_at = time.time()
+            session.commit()
+        except Exception as e:
+            session.rollback()
+            logger.error(f"reset_user_password error: {e}")
+        finally:
+            session.close()
+
+    def get_user_prompt(self, user_id: str, key: str) -> Optional[str]:
+        if not user_id or not key:
+            return None
+        session = self._get_session()
+        try:
+            p = session.query(UserPrompt).filter_by(user_id=user_id, key=key).first()
+            return p.value if p else None
+        finally:
+            session.close()
+
+    def set_user_prompt(self, user_id: str, key: str, value: Any) -> None:
+        if not user_id or not key:
+            return
+        session = self._get_session()
+        try:
+            now = time.time()
+            v = json.dumps(value, ensure_ascii=False) if not isinstance(value, str) else value
+            p = session.query(UserPrompt).filter_by(user_id=user_id, key=key).first()
+            if p:
+                p.value = v
+                p.updated_at = now
+            else:
+                p = UserPrompt(id=secrets.token_hex(16), user_id=user_id, key=key, value=v, updated_at=now)
+                session.add(p)
+            session.commit()
+        except Exception as e:
+            session.rollback()
+            logger.error(f"set_user_prompt error: {e}")
+        finally:
+            session.close()
+
+    # --- RBAC helpers ---
+    def list_projects_for_user(self, user: Dict[str, Any]) -> List[Dict[str, Any]]:
+        if not user:
+            return []
+        if user.get("role") == "admin":
+            return self.list_projects()
+        uid = user.get("id")
+        session = self._get_session()
+        try:
+            projects = session.query(Project).filter_by(owner_user_id=uid).order_by(Project.updated_at.desc()).all()
+            return [{"id": p.id, "name": p.name, "status": p.status, "owner_user_id": p.owner_user_id, "updated_at": p.updated_at} for p in projects]
+        finally:
+            session.close()
+
+    def get_project_for_user(self, project_id: str, user: Dict[str, Any]) -> Optional[Dict[str, Any]]:
+        data = self.get_project(project_id)
+        if not data or not user:
+            return None
+        if user.get("role") == "admin":
+            return data
+        if data.get("owner_user_id") != user.get("id"):
+            return None
+        return data
+
+    def migrate_projects_owner_to(self, owner_user_id: str) -> int:
+        """Assign owner_user_id for legacy projects where it is NULL/empty."""
+        session = self._get_session()
+        try:
+            q = session.query(Project).filter((Project.owner_user_id == None) | (Project.owner_user_id == ""))  # noqa: E711
+            rows = q.all()
+            for p in rows:
+                p.owner_user_id = owner_user_id
+                p.updated_at = time.time()
+            session.commit()
+            return len(rows)
+        except Exception as e:
+            session.rollback()
+            logger.error(f"migrate_projects_owner_to error: {e}")
+            return 0
+        finally:
+            session.close()
+
     # --- Asset/Task Operations ---
 
     def save_asset(self, project_id: str, scene_id: int, asset_type: str, 
@@ -253,6 +659,92 @@ class DBManager:
         finally:
             session.close()
 
+    # --- Render Job Operations ---
+
+    def create_render_job(
+        self,
+        job_id: str,
+        project_id: str,
+        *,
+        status: str = "queued",
+        progress: float = 0.0,
+        message: str = "",
+        request: Optional[Dict[str, Any]] = None,
+        parent_id: Optional[str] = None,
+    ) -> None:
+        session = self._get_session()
+        try:
+            existing = session.query(RenderJob).filter_by(id=job_id).first()
+            if existing:
+                return
+            now = time.time()
+            rj = RenderJob(
+                id=job_id,
+                project_id=project_id,
+                status=status,
+                progress=float(progress or 0.0),
+                message=message or "",
+                request_json=json.dumps(request, ensure_ascii=False) if request is not None else None,
+                parent_id=parent_id,
+                created_at=now,
+                updated_at=now,
+            )
+            session.add(rj)
+            session.commit()
+        except Exception as e:
+            session.rollback()
+            logger.error(f"create_render_job error: {e}")
+            raise
+        finally:
+            session.close()
+
+    def update_render_job(self, job_id: str, patch: Dict[str, Any]) -> None:
+        if not job_id or not patch:
+            return
+        session = self._get_session()
+        try:
+            rj = session.query(RenderJob).filter_by(id=job_id).first()
+            if not rj:
+                return
+            for k, v in patch.items():
+                if not hasattr(rj, k):
+                    continue
+                setattr(rj, k, v)
+            rj.updated_at = time.time()
+            session.commit()
+        except Exception as e:
+            session.rollback()
+            logger.error(f"update_render_job error: {e}")
+        finally:
+            session.close()
+
+    def get_render_job(self, job_id: str) -> Optional[Dict[str, Any]]:
+        session = self._get_session()
+        try:
+            rj = session.query(RenderJob).filter_by(id=job_id).first()
+            if not rj:
+                return None
+            try:
+                req = json.loads(rj.request_json) if rj.request_json else None
+            except Exception:
+                req = None
+            return {
+                "id": rj.id,
+                "project_id": rj.project_id,
+                "status": rj.status,
+                "progress": float(rj.progress or 0.0),
+                "message": rj.message or "",
+                "output_path": rj.output_path,
+                "output_url": rj.output_url,
+                "error": rj.error,
+                "request": req,
+                "parent_id": rj.parent_id,
+                "created_at": rj.created_at,
+                "updated_at": rj.updated_at,
+            }
+        finally:
+            session.close()
+
     def get_asset(self, project_id: str, scene_id: int, asset_type: str) -> Optional[Dict[str, Any]]:
         session = self._get_session()
         try:
@@ -279,6 +771,30 @@ class DBManager:
         finally:
             session.close()
 
+    def get_asset_by_id(self, asset_id: int) -> Optional[Dict[str, Any]]:
+        """通过自增 id 获取素材记录（用于 file proxy）。"""
+        session = self._get_session()
+        try:
+            a = session.query(SceneAsset).filter_by(id=int(asset_id)).first()
+            if not a:
+                return None
+            return {
+                "id": a.id,
+                "project_id": a.project_id,
+                "scene_id": a.scene_id,
+                "asset_type": a.asset_type,
+                "status": a.status,
+                "local_path": a.local_path,
+                "remote_url": a.remote_url,
+                "task_id": a.task_id,
+                "metadata": json.loads(a.metadata_json) if a.metadata_json else {},
+                "updated_at": a.updated_at,
+            }
+        except Exception:
+            return None
+        finally:
+            session.close()
+
     def update_asset_metadata(self, project_id: str, scene_id: int, asset_type: str, patch: Dict[str, Any]) -> None:
         """Merge-patch asset.metadata JSON without overwriting other fields."""
         if not patch:

modules/ffmpeg_utils.py

@@ -172,7 +172,8 @@ def get_video_info(video_path: str) -> Dict[str, Any]:
 def concat_videos(
     video_paths: List[str],
     output_path: str,
-    target_size: Tuple[int, int] = (1080, 1920)
+    target_size: Tuple[int, int] = (1080, 1920),
+    fades: Optional[List[Dict[str, float]]] = None
 ) -> str:
     """
     使用 FFmpeg concat demuxer 拼接多段视频
@@ -197,10 +198,72 @@ def concat_videos(
     filter_parts = []
     for i in range(len(video_paths)):
         # scale 保持宽高比，pad 填充黑边居中
-        filter_parts.append(
+        chain = (
             f"[{i}:v]scale={width}:{height}:force_original_aspect_ratio=decrease,"
-            f"pad={width}:{height}:(ow-iw)/2:(oh-ih)/2:black,setsar=1[v{i}]"
+            f"pad={width}:{height}:(ow-iw)/2:(oh-ih)/2:black,setsar=1"
         )
+        # 可选：片段末尾“火山式转场”（不改时长、不重叠）
+        if fades and i < len(fades):
+            fx = fades[i] or {}
+            fi = float(fx.get("in", 0) or 0.0)
+            fo = float(fx.get("out", 0) or 0.0)
+            t_type = str(fx.get("type") or "")
+            t_dur = float(fx.get("dur") or 0.0)
+
+            try:
+                dur = float(get_video_info(video_paths[i]).get("duration") or 0.0)
+            except Exception:
+                dur = 0.0
+
+            # 基础淡入/淡出
+            if fi > 0:
+                chain += f",fade=t=in:st=0:d={fi}"
+            if fo > 0 and dur > 0:
+                st = max(dur - fo, 0.0)
+                chain += f",fade=t=out:st={st}:d={fo}"
+
+            # 末尾动效（WYSIWYG：前端预览必须与此一致）
+            if t_type and t_dur > 0 and dur > 0:
+                st = max(dur - t_dur, 0.0)
+                td = max(t_dur, 0.001)
+                p = f"if(between(t\\,{st}\\,{dur})\\,(t-{st})/{td}\\,0)"
+                if t_type == "fade":
+                    chain += f",fade=t=out:st={st}:d={t_dur}"
+                elif t_type == "fadeWhite":
+                    chain += f",fade=t=out:st={st}:d={t_dur}:color=white"
+                elif t_type == "blurOut":
+                    chain += f",gblur=sigma='10*{p}':steps=1"
+                elif t_type == "blurFade":
+                    chain += f",gblur=sigma='8*{p}':steps=1,fade=t=out:st={st}:d={t_dur}"
+                elif t_type == "flash":
+                    chain += f",eq=brightness='0.7*(1-abs(0.5-{p})*2)'"
+                elif t_type == "desaturate":
+                    chain += f",hue=s='1-0.9*{p}'"
+                elif t_type == "colorPop":
+                    chain += f",hue=s='1+0.8*{p}',eq=contrast='1+0.3*{p}'"
+                elif t_type == "hueShift":
+                    chain += f",hue=h='60*{p}'"
+                elif t_type == "darken":
+                    chain += f",eq=brightness='-0.4*{p}'"
+                elif t_type in ("slideLeft", "slideRight", "slideUp", "slideDown"):
+                    off = 80
+                    if t_type == "slideLeft":
+                        chain += f",pad={width+off}:{height}:{off/2}-{off}*{p}:0:black,crop={width}:{height}:{off/2}:0"
+                    if t_type == "slideRight":
+                        chain += f",pad={width+off}:{height}:{off/2}+{off}*{p}:0:black,crop={width}:{height}:{off/2}:0"
+                    if t_type == "slideUp":
+                        chain += f",pad={width}:{height+off}:0:{off/2}-{off}*{p}:black,crop={width}:{height}:0:{off/2}"
+                    if t_type == "slideDown":
+                        chain += f",pad={width}:{height+off}:0:{off/2}+{off}*{p}:black,crop={width}:{height}:0:{off/2}"
+                elif t_type in ("zoomOut", "zoomIn"):
+                    if t_type == "zoomOut":
+                        chain += f",scale=w='{width}*(1-0.10*{p})':h='{height}*(1-0.10*{p})':eval=frame,pad={width}:{height}:(ow-iw)/2:(oh-ih)/2:black"
+                    else:
+                        chain += f",scale=w='{width}*(1+0.10*{p})':h='{height}*(1+0.10*{p})':eval=frame,crop={width}:{height}"
+                elif t_type == "rotateOut":
+                    chain += f",rotate=a='0.12*{p}':c=black@1:ow={width}:oh={height}"
+        chain += f"[v{i}]"
+        filter_parts.append(chain)
     
     # 拼接所有视频流
     concat_inputs = "".join([f"[v{i}]" for i in range(len(video_paths))])
@@ -232,7 +295,8 @@ def concat_videos(
 def concat_videos_with_audio(
     video_paths: List[str],
     output_path: str,
-    target_size: Tuple[int, int] = (1080, 1920)
+    target_size: Tuple[int, int] = (1080, 1920),
+    fades: Optional[List[Dict[str, float]]] = None
 ) -> str:
     """
     拼接视频并保留音频轨道
@@ -250,10 +314,70 @@ def concat_videos_with_audio(
     
     # 视频处理
     for i in range(n):
-        filter_parts.append(
+        chain = (
             f"[{i}:v]scale={width}:{height}:force_original_aspect_ratio=decrease,"
-            f"pad={width}:{height}:(ow-iw)/2:(oh-ih)/2:black,setsar=1[v{i}]"
+            f"pad={width}:{height}:(ow-iw)/2:(oh-ih)/2:black,setsar=1"
         )
+        # 可选：片段末尾“火山式转场”（不改时长、不重叠）
+        if fades and i < len(fades):
+            fx = fades[i] or {}
+            fi = float(fx.get("in", 0) or 0.0)
+            fo = float(fx.get("out", 0) or 0.0)
+            t_type = str(fx.get("type") or "")
+            t_dur = float(fx.get("dur") or 0.0)
+
+            try:
+                dur = float(get_video_info(video_paths[i]).get("duration") or 0.0)
+            except Exception:
+                dur = 0.0
+
+            if fi > 0:
+                chain += f",fade=t=in:st=0:d={fi}"
+            if fo > 0 and dur > 0:
+                st = max(dur - fo, 0.0)
+                chain += f",fade=t=out:st={st}:d={fo}"
+
+            if t_type and t_dur > 0 and dur > 0:
+                st = max(dur - t_dur, 0.0)
+                td = max(t_dur, 0.001)
+                p = f"if(between(t\\,{st}\\,{dur})\\,(t-{st})/{td}\\,0)"
+                if t_type == "fade":
+                    chain += f",fade=t=out:st={st}:d={t_dur}"
+                elif t_type == "fadeWhite":
+                    chain += f",fade=t=out:st={st}:d={t_dur}:color=white"
+                elif t_type == "blurOut":
+                    chain += f",gblur=sigma='10*{p}':steps=1"
+                elif t_type == "blurFade":
+                    chain += f",gblur=sigma='8*{p}':steps=1,fade=t=out:st={st}:d={t_dur}"
+                elif t_type == "flash":
+                    chain += f",eq=brightness='0.7*(1-abs(0.5-{p})*2)'"
+                elif t_type == "desaturate":
+                    chain += f",hue=s='1-0.9*{p}'"
+                elif t_type == "colorPop":
+                    chain += f",hue=s='1+0.8*{p}',eq=contrast='1+0.3*{p}'"
+                elif t_type == "hueShift":
+                    chain += f",hue=h='60*{p}'"
+                elif t_type == "darken":
+                    chain += f",eq=brightness='-0.4*{p}'"
+                elif t_type in ("slideLeft", "slideRight", "slideUp", "slideDown"):
+                    off = 80
+                    if t_type == "slideLeft":
+                        chain += f",pad={width+off}:{height}:{off/2}-{off}*{p}:0:black,crop={width}:{height}:{off/2}:0"
+                    if t_type == "slideRight":
+                        chain += f",pad={width+off}:{height}:{off/2}+{off}*{p}:0:black,crop={width}:{height}:{off/2}:0"
+                    if t_type == "slideUp":
+                        chain += f",pad={width}:{height+off}:0:{off/2}-{off}*{p}:black,crop={width}:{height}:0:{off/2}"
+                    if t_type == "slideDown":
+                        chain += f",pad={width}:{height+off}:0:{off/2}+{off}*{p}:black,crop={width}:{height}:0:{off/2}"
+                elif t_type in ("zoomOut", "zoomIn"):
+                    if t_type == "zoomOut":
+                        chain += f",scale=w='{width}*(1-0.10*{p})':h='{height}*(1-0.10*{p})':eval=frame,pad={width}:{height}:(ow-iw)/2:(oh-ih)/2:black"
+                    else:
+                        chain += f",scale=w='{width}*(1+0.10*{p})':h='{height}*(1+0.10*{p})':eval=frame,crop={width}:{height}"
+                elif t_type == "rotateOut":
+                    chain += f",rotate=a='0.12*{p}':c=black@1:ow={width}:oh={height}"
+        chain += f"[v{i}]"
+        filter_parts.append(chain)
     
     # 音频处理（静音填充如果没有音频）
     for i in range(n):
@@ -469,6 +593,127 @@ def adjust_audio_duration(
     return output_path
 
 
+def _atempo_chain(speed: float) -> str:
+    """
+    构造 atempo 链，支持 <0.5 或 >2.0 的倍速（通过链式 atempo）。
+    """
+    try:
+        s = float(speed)
+    except Exception:
+        s = 1.0
+    if s <= 0:
+        s = 1.0
+    parts = []
+    # atempo 支持 0.5~2.0
+    while s > 2.0:
+        parts.append("atempo=2.0")
+        s /= 2.0
+    while s < 0.5:
+        parts.append("atempo=0.5")
+        s /= 0.5
+    parts.append(f"atempo={s}")
+    return ",".join(parts)
+
+
+def change_audio_speed(input_path: str, speed: float, output_path: str) -> str:
+    """改变音频播放倍速（纯播放倍速）。"""
+    if not os.path.exists(input_path):
+        return None
+    af = _atempo_chain(speed)
+    cmd = [FFMPEG_PATH, "-y", "-i", input_path, "-filter:a", af, output_path]
+    _run_ffmpeg(cmd)
+    return output_path
+
+
+def fit_audio_to_duration_by_speed(input_path: str, target_duration: float, output_path: str) -> str:
+    """
+    通过“改变播放倍速”来贴合目标时长（可快可慢），并裁剪/补齐到严格时长。
+    适用于旁白：用户拉伸片段期望语速变化，而不是静音补齐。
+    """
+    if not os.path.exists(input_path):
+        return None
+    try:
+        td = float(target_duration or 0)
+    except Exception:
+        td = 0.0
+    if td <= 0:
+        import shutil
+        shutil.copy(input_path, output_path)
+        return output_path
+
+    cur = float(get_audio_info(input_path).get("duration") or 0.0)
+    if cur <= 0:
+        import shutil
+        shutil.copy(input_path, output_path)
+        return output_path
+
+    speed = cur / td
+    af_speed = _atempo_chain(speed)
+    # 贴合后仍做一次 atrim+apad 保证严格时长（避免累计误差）
+    af = f"{af_speed},atrim=0:{td},apad=pad_dur=0,atrim=0:{td}"
+    cmd = [FFMPEG_PATH, "-y", "-i", input_path, "-filter:a", af, output_path]
+    _run_ffmpeg(cmd)
+    return output_path
+
+
+def force_audio_duration(input_path: str, target_duration: float, output_path: str) -> str:
+    """不改变倍速，仅裁剪/补齐到严格时长（用于倍速已在上游完成的场景）。"""
+    if not os.path.exists(input_path):
+        return None
+    try:
+        td = float(target_duration or 0)
+    except Exception:
+        td = 0.0
+    if td <= 0:
+        import shutil
+        shutil.copy(input_path, output_path)
+        return output_path
+    af = f"atrim=0:{td},apad=pad_dur=0,atrim=0:{td}"
+    cmd = [FFMPEG_PATH, "-y", "-i", input_path, "-filter:a", af, output_path]
+    _run_ffmpeg(cmd)
+    return output_path
+
+
+def _which(cmd: str) -> Optional[str]:
+    try:
+        import shutil
+        return shutil.which(cmd)
+    except Exception:
+        return None
+
+
+def normalize_sticker_to_png(input_path: str, output_path: str) -> str:
+    """
+    将贴纸规范化为 PNG（用于 ffmpeg overlay）。
+    - PNG/WEBP：直接返回原图或拷贝
+    - SVG：优先用 rsvg-convert 转 PNG；否则尝试 ffmpeg 直接解码
+    """
+    if not input_path or not os.path.exists(input_path):
+        return None
+    ext = Path(input_path).suffix.lower()
+    if ext in [".png"]:
+        return input_path
+    if ext in [".webp"]:
+        # 转 PNG，避免某些 ffmpeg build 对 webp 支持不一致
+        cmd = [FFMPEG_PATH, "-y", "-i", input_path, output_path]
+        _run_ffmpeg(cmd)
+        return output_path
+    if ext == ".svg":
+        rsvg = _which("rsvg-convert")
+        if rsvg:
+            import subprocess
+            subprocess.check_call([rsvg, "-o", output_path, input_path])
+            return output_path
+        # fallback: ffmpeg decode svg（依赖 build）
+        cmd = [FFMPEG_PATH, "-y", "-i", input_path, output_path]
+        _run_ffmpeg(cmd)
+        return output_path
+    # 其他格式：尽量用 ffmpeg 转
+    cmd = [FFMPEG_PATH, "-y", "-i", input_path, output_path]
+    _run_ffmpeg(cmd)
+    return output_path
+
+
 def get_audio_info(file_path: str) -> Dict[str, Any]:
     """获取音频信息"""
     return get_video_info(file_path)
@@ -830,6 +1075,12 @@ def add_bgm(
     loop: bool = True,
     ducking: bool = True,
     duck_gain_db: float = -6.0,
+    # 新增：按时间段闪避（更可控，和旁白时间轴严格对齐）
+    duck_volume: float = 0.25,
+    duck_ranges: Optional[List[Tuple[float, float]]] = None,
+    # 新增：BGM 片段可有起点/时长（不强制从 0 覆盖整段视频）
+    start_time: float = 0.0,
+    clip_duration: Optional[float] = None,
     fade_in: float = 1.0,
     fade_out: float = 1.0
 ) -> str:
@@ -856,30 +1107,46 @@ def add_bgm(
     info = get_video_info(video_path)
     video_duration = info["duration"]
     
-    if loop:
+    # 片段时长：默认覆盖整段视频
-        bgm_chain = (
+    dur = float(clip_duration) if (clip_duration is not None and float(clip_duration) > 0) else float(video_duration)
-            f"[1:a]aloop=-1:size=2e+09,asetpts=N/SR/TB,"
+    st = max(0.0, float(start_time or 0.0))
-            f"atrim=0:{video_duration},"
+    end_for_fade = max(dur - float(fade_out or 0.0), 0.0)
-            f"afade=t=in:st=0:d={fade_in},"
-            f"afade=t=out:st={max(video_duration - fade_out, 0)}:d={fade_out},"
-            f"volume={bgm_volume}[bgm]"
-        )
-    else:
-        bgm_chain = (
-            f"[1:a]"
-            f"afade=t=in:st=0:d={fade_in},"
-            f"afade=t=out:st={max(video_duration - fade_out, 0)}:d={fade_out},"
-            f"volume={bgm_volume}[bgm]"
-        )
 
-    if ducking:
+    # 基础链：loop/trim -> fades -> base volume
-        # 使用安全参数的 sidechaincompress，避免 unsupported 参数
+    if loop:
+        bgm_chain = f"[1:a]aloop=-1:size=2e+09,asetpts=N/SR/TB,atrim=0:{dur}"
+    else:
+        bgm_chain = f"[1:a]atrim=0:{dur}"
+
+    bgm_chain += f",afade=t=in:st=0:d={float(fade_in or 0.0)},afade=t=out:st={end_for_fade}:d={float(fade_out or 0.0)},volume={bgm_volume}"
+
+    # 延迟到 start_time
+    if st > 1e-6:
+        ms = int(st * 1000)
+        bgm_chain += f",adelay={ms}|{ms}"
+
+    # 闪避（按时间段）
+    # 注意：使用 enable 让 filter 只在区间内生效（外部直接 passthrough）
+    if ducking and duck_ranges:
+        dv = max(0.05, min(1.0, float(duck_volume or 0.25)))
+        for (rs, re) in duck_ranges:
+            rsf = max(0.0, float(rs))
+            ref = max(rsf, float(re))
+            bgm_chain += f",volume={dv}:enable='between(t,{rsf},{ref})'"
+
+    bgm_chain += "[bgm]"
+
+    # 如果提供了 duck_ranges，就用确定性的 amix（ducking 已在 bgm_chain 内完成）
+    if ducking and duck_ranges:
+        filter_complex = f"{bgm_chain};[0:a][bgm]amix=inputs=2:duration=first:dropout_transition=0:normalize=0[outa]"
+    elif ducking:
+        # 否则退回 sidechaincompress（对原视频音频进行侧链压缩）
         filter_complex = (
             f"{bgm_chain};"
             f"[0:a][bgm]sidechaincompress=threshold=0.1:ratio=4:attack=5:release=250:makeup=1:mix=1:level_in=1:level_sc=1[outa]"
         )
     else:
-        filter_complex = f"{bgm_chain};[0:a][bgm]amix=inputs=2:duration=first[outa]"
+        filter_complex = f"{bgm_chain};[0:a][bgm]amix=inputs=2:duration=first:dropout_transition=0:normalize=0[outa]"
     
     cmd = [
         FFMPEG_PATH, "-y",

modules/legacy_normalizer.py

@@ -246,3 +246,9 @@ def normalize_legacy_project(doc: Dict[str, Any]) -> Dict[str, Any]:
 
 
 
+
+
+
+
+
+

modules/legacy_path_mapper.py

@@ -19,6 +19,7 @@ from typing import Optional, Tuple
 
 LEGACY_HOST_TEMP_PREFIX = "/root/video-flow/temp/"
 LEGACY_HOST_OUTPUT_PREFIX = "/root/video-flow/output/"
+LEGACY_HOST_PREFIX = "/root/video-flow/"
 
 # Container mount points (see docker-compose.yml)
 LEGACY_CONTAINER_TEMP_DIR = "/legacy/temp"
@@ -42,18 +43,27 @@ def map_legacy_local_path(local_path: Optional[str]) -> Tuple[Optional[str], Opt
     if os.path.exists(local_path):
         return local_path, None
 
-    # Legacy host -> container mapping by basename
+    # Legacy host path -> current container workspace path (same repo but different prefix)
+    # Example:
+    #   /root/video-flow/temp/projects/...  ->  /app/temp/projects/...
+    # This covers cases where we don't mount /legacy/* but the files were copied into current stack.
+    if local_path.startswith(LEGACY_HOST_PREFIX):
+        rest = local_path[len(LEGACY_HOST_PREFIX):].lstrip("/")
+        candidate = str(Path("/app") / rest)
+        if os.path.exists(candidate):
+            return candidate, None
+
+    # Legacy host -> container mapping (preserve relative path)
     if local_path.startswith(LEGACY_HOST_TEMP_PREFIX):
-        name = Path(local_path).name
+        rel = local_path[len(LEGACY_HOST_TEMP_PREFIX):].lstrip("/")
-        container_path = str(Path(LEGACY_CONTAINER_TEMP_DIR) / name)
+        container_path = str(Path(LEGACY_CONTAINER_TEMP_DIR) / rel)
-        url = f"{LEGACY_STATIC_TEMP_PREFIX}{name}"
+        # 静态路由通常只覆盖目录根（不包含子目录）；这里交给 /api/assets/file 做 FileResponse 更稳
-        return container_path, url
+        return container_path, None
 
     if local_path.startswith(LEGACY_HOST_OUTPUT_PREFIX):
-        name = Path(local_path).name
+        rel = local_path[len(LEGACY_HOST_OUTPUT_PREFIX):].lstrip("/")
-        container_path = str(Path(LEGACY_CONTAINER_OUTPUT_DIR) / name)
+        container_path = str(Path(LEGACY_CONTAINER_OUTPUT_DIR) / rel)
-        url = f"{LEGACY_STATIC_OUTPUT_PREFIX}{name}"
+        return container_path, None
-        return container_path, url
 
     # Unknown path: keep as-is
     return local_path, None
@@ -64,3 +74,9 @@ def map_legacy_local_path(local_path: Optional[str]) -> Tuple[Optional[str], Opt
 
 
 
+
+
+
+
+
+

modules/preview_proxy.py

@@ -0,0 +1,102 @@
+"""
+Generate and cache low-bitrate preview proxies for browser playback.
+
+Goal:
+- Improve Remotion Player preview smoothness by serving smaller/faster-to-decode videos.
+- Keep original `source_path` for accurate export; only swap `source_url` for preview.
+
+Design:
+- Deterministic cache key based on (path, mtime, size).
+- Proxy lives under config.TEMP_DIR / "proxy".
+- Generated with ffmpeg: scale/pad + fps downsample + faststart + no audio.
+"""
+
+from __future__ import annotations
+
+import hashlib
+import os
+from pathlib import Path
+from typing import Optional, Tuple
+
+import config
+from modules import ffmpeg_utils
+
+
+def _key_for_file(path: str) -> str:
+    st = os.stat(path)
+    raw = f"{path}|{st.st_mtime_ns}|{st.st_size}".encode("utf-8")
+    return hashlib.sha1(raw).hexdigest()  # short, deterministic
+
+
+def ensure_video_proxy(
+    source_path: str,
+    *,
+    target_w: int = 540,
+    target_h: int = 960,
+    target_fps: int = 24,
+    crf: int = 28,
+    preset: str = "veryfast",
+) -> Tuple[Optional[str], Optional[str]]:
+    """
+    Ensure a preview proxy exists for source_path.
+
+    Returns: (proxy_path, proxy_url) or (None, None) if source_path invalid.
+    """
+    if not source_path or not os.path.exists(source_path):
+        return None, None
+
+    proxy_dir = Path(config.TEMP_DIR) / "proxy"
+    proxy_dir.mkdir(parents=True, exist_ok=True)
+
+    key = _key_for_file(source_path)
+    out_name = f"proxy_{key}.mp4"
+    out_path = proxy_dir / out_name
+
+    if out_path.exists() and out_path.stat().st_size > 1024:
+        return str(out_path), f"/static/temp/proxy/{out_name}"
+
+    vf = (
+        f"scale={target_w}:{target_h}:force_original_aspect_ratio=decrease,"
+        f"pad={target_w}:{target_h}:(ow-iw)/2:(oh-ih)/2:black,"
+        f"fps={target_fps}"
+    )
+
+    cmd = [
+        ffmpeg_utils.FFMPEG_PATH,
+        "-y",
+        "-i",
+        source_path,
+        "-an",  # preview 不要音轨，减少解码负担（旁白/BGM 走单独轨道）
+        "-vf",
+        vf,
+        "-c:v",
+        "libx264",
+        "-preset",
+        preset,
+        "-crf",
+        str(crf),
+        "-tune",
+        "fastdecode",
+        "-pix_fmt",
+        "yuv420p",
+        "-movflags",
+        "+faststart",
+        str(out_path),
+    ]
+
+    ffmpeg_utils._run_ffmpeg(cmd)
+    if out_path.exists() and out_path.stat().st_size > 1024:
+        return str(out_path), f"/static/temp/proxy/{out_name}"
+    return None, None
+
+
+
+
+
+
+
+
+
+
+
+

modules/script_gen.py

@@ -6,6 +6,7 @@ import base64
 import json
 import logging
 import os
+import time
 import requests
 from typing import Dict, Any, List, Optional
 from pathlib import Path
@@ -27,7 +28,10 @@ class ScriptGenerator:
         # OpenAI-compatible client for ShuBiaoBiao (supports multiple models incl. GPT)
         self.shubiaobiao_client = OpenAI(
             api_key=config.SHUBIAOBIAO_KEY,
-            base_url=config.SHUBIAOBIAO_BASE_URL
+            base_url=config.SHUBIAOBIAO_BASE_URL,
+            # IMPORTANT: OpenAI SDK default timeout is 10 minutes; cap it to keep UX responsive.
+            timeout=config.SHUBIAOBIAO_CHAT_TIMEOUT_S,
+            max_retries=config.SHUBIAOBIAO_CHAT_MAX_RETRIES,
         )
         
         # Default System Prompt
@@ -139,15 +143,23 @@ class ScriptGenerator:
         product_name: str, 
         product_info: Dict[str, Any], 
         image_paths: List[str] = None,
-        model_provider: str = "shubiaobiao" # "shubiaobiao" or "doubao"
+        model_provider: str = "shubiaobiao", # "shubiaobiao" or "doubao"
+        user_id: Optional[str] = None,
     ) -> Dict[str, Any]:
         """
         生成分镜脚本
         """
         logger.info(f"Generating script for: {product_name} (Provider: {model_provider})")
         
-        # 1. 构造 Prompt (优先从数据库读取配置)
+        # 1. 构造 Prompt (优先按 user_id 读取；否则回退到全局配置，再回退默认)
-        system_prompt = db.get_config("prompt_script_gen", self.default_system_prompt)
+        system_prompt = None
+        if user_id:
+            try:
+                system_prompt = db.get_user_prompt(user_id, "prompt_script_gen")
+            except Exception:
+                system_prompt = None
+        if not system_prompt:
+            system_prompt = db.get_config("prompt_script_gen", self.default_system_prompt)
         user_prompt = self._build_user_prompt(product_name, product_info)
         
         # Branch for Doubao (Volcengine)
@@ -293,21 +305,40 @@ class ScriptGenerator:
         ShuBiaoBiao OpenAI-compatible multimodal chat.
         IMPORTANT: For ShuBiaoBiao models, we pass image URLs (R2 public URLs), not base64.
         """
+        t0 = time.time()
+        # Use WARNING level so it shows up even if Streamlit/root logger is not configured to INFO.
+        logger.warning(
+            f"[script_gen] start shubiaobiao chat model={model_name} images={len(image_paths or [])} "
+            f"timeout_s={getattr(config, 'SHUBIAOBIAO_CHAT_TIMEOUT_S', 'n/a')} "
+            f"max_retries={getattr(config, 'SHUBIAOBIAO_CHAT_MAX_RETRIES', 'n/a')}"
+        )
         messages = [{"role": "system", "content": system_prompt}]
         user_content: List[Dict[str, Any]] = []
         # Images first (URL), then text
+        t_upload0 = time.time()
         urls = self._upload_images_to_r2(image_paths or [], limit=10)
+        logger.warning(
+            f"[script_gen] r2_upload done urls={len(urls)} elapsed_s={time.time() - t_upload0:.2f}"
+        )
         for url in urls:
             user_content.append({"type": "image_url", "image_url": {"url": url}})
         user_content.append({"type": "text", "text": user_prompt})
         messages.append({"role": "user", "content": user_content})
 
         try:
-            resp = self.shubiaobiao_client.chat.completions.create(
+            client = self.shubiaobiao_client.with_options(
+                timeout=config.SHUBIAOBIAO_CHAT_TIMEOUT_S,
+                max_retries=config.SHUBIAOBIAO_CHAT_MAX_RETRIES,
+            )
+            t_call0 = time.time()
+            resp = client.chat.completions.create(
                 model=model_name,
                 messages=messages,
                 temperature=0.7,
             )
+            logger.warning(
+                f"[script_gen] shubiaobiao chat done elapsed_s={time.time() - t_call0:.2f} total_s={time.time() - t0:.2f}"
+            )
             content_text = (resp.choices[0].message.content or "").strip()
             script_json = self._extract_json_from_response(content_text)
             if script_json is None:
@@ -323,7 +354,9 @@ class ScriptGenerator:
             }
             return final_script
         except Exception as e:
-            logger.error(f"shubiaobiao script generation failed ({model_name}): {e}")
+            logger.error(
+                f"shubiaobiao script generation failed ({model_name}) after {time.time() - t0:.2f}s: {e}"
+            )
             return None
 
     def _postprocess_selling_points(self, product_info: Dict[str, Any], selling_points: Any) -> List[str]:
@@ -582,7 +615,33 @@ class ScriptGenerator:
 
     def _validate_and_fix_script(self, script: Dict[str, Any]) -> Dict[str, Any]:
         """校验并修复脚本结构"""
-        # 简单校验，确保必要字段存在
+        if not isinstance(script, dict):
-        if "scenes" not in script:
+            return {"scenes": []}
+
+        # Ensure fields exist
+        if "scenes" not in script or not isinstance(script.get("scenes"), list):
             script["scenes"] = []
+
+        # Normalize: keep visual_anchor at top-level, but avoid repeating the full anchor in every scene.visual_prompt.
+        # Reason: repeating a long anchor 4-5 times explodes tokens and makes UI look like "only three sections",
+        # while image generation already supports passing visual_anchor separately and prepending it at runtime.
+        visual_anchor = script.get("visual_anchor") or ""
+        if isinstance(visual_anchor, str) and visual_anchor.strip() and script["scenes"]:
+            anchor = visual_anchor.strip()
+            prefix = f"[{anchor}]"
+            for scene in script["scenes"]:
+                if not isinstance(scene, dict):
+                    continue
+                vp = scene.get("visual_prompt")
+                if not isinstance(vp, str) or not vp.strip():
+                    continue
+                s = vp.strip()
+                # Strip exact "[anchor]" prefix if present
+                if s.startswith(prefix):
+                    s = s[len(prefix):].lstrip()
+                # If the model output copied the raw anchor without brackets, strip it too
+                elif s.startswith(anchor):
+                    s = s[len(anchor):].lstrip()
+                scene["visual_prompt"] = s
+
         return script

modules/text_renderer.py

@@ -90,6 +90,79 @@ class TextRenderer:
             return color
         return (0, 0, 0, 255)
 
+    def _wrap_text_to_width(self, text: str, font: ImageFont.FreeTypeFont, max_width: int) -> str:
+        """
+        将文本按最大宽度自动换行（支持中英文混排）。
+        - 保留原始换行符为段落边界
+        - 英文优先按空格断词；中文按字符贪心换行
+        """
+        try:
+            mw = int(max_width or 0)
+        except Exception:
+            mw = 0
+        if mw <= 0:
+            return text
+
+        # 兼容：去掉末尾多余空行
+        raw_paras = (text or "").split("\n")
+        out_lines: List[str] = []
+
+        # 1x1 dummy draw 用于测量
+        dummy_draw = ImageDraw.Draw(Image.new("RGBA", (1, 1)))
+
+        def text_w(s: str) -> float:
+            try:
+                return float(dummy_draw.textlength(s, font=font))
+            except Exception:
+                bbox = dummy_draw.textbbox((0, 0), s, font=font)
+                return float((bbox[2] - bbox[0]) if bbox else 0)
+
+        for para in raw_paras:
+            p = (para or "").rstrip()
+            if not p:
+                out_lines.append("")
+                continue
+
+            # 英文/混排：尝试按空格分词，否则按字符
+            use_words = (" " in p)
+            tokens = p.split(" ") if use_words else list(p)
+
+            cur = ""
+            for tok in tokens:
+                cand = (cur + (" " if (use_words and cur) else "") + tok) if cur else tok
+                if text_w(cand) <= mw:
+                    cur = cand
+                    continue
+
+                # 当前行放不下：先落一行
+                if cur:
+                    out_lines.append(cur)
+                    cur = tok
+                else:
+                    # 单 token 超宽：强制按字符拆
+                    if use_words:
+                        chars = list(tok)
+                    else:
+                        chars = [tok]
+                    buf = ""
+                    for ch in chars:
+                        cand2 = buf + ch
+                        if text_w(cand2) <= mw or not buf:
+                            buf = cand2
+                        else:
+                            out_lines.append(buf)
+                            buf = ch
+                    cur = buf
+
+            if cur:
+                out_lines.append(cur)
+
+        # 去掉尾部空行（保持中间空行）
+        while out_lines and out_lines[-1] == "":
+            out_lines.pop()
+
+        return "\n".join(out_lines)
+
     def render(self, text: str, style: Union[Dict[str, Any], str], cache: bool = True) -> str:
         """
         渲染文本并返回图片路径
@@ -122,14 +195,29 @@ class TextRenderer:
         font_size = style.get("font_size", 60)
         font = self._get_font(font_path, font_size)
         font_color = self._parse_color(style.get("font_color", "#FFFFFF"))
+        bold = bool(style.get("bold", False))
+        italic = bool(style.get("italic", False))
+        underline = bool(style.get("underline", False))
         
-        # 3. 测量文本尺寸
+        # 3. 自动换行（可选）
+        max_width = style.get("max_width") or style.get("maxWidth") or style.get("text_box_width")
+        try:
+            max_width = int(max_width) if max_width is not None else 0
+        except Exception:
+            max_width = 0
+        if max_width > 0:
+            text = self._wrap_text_to_width(text, font, max_width)
+
+        # 4. 测量文本尺寸（支持多行）
         dummy_draw = ImageDraw.Draw(Image.new("RGBA", (1, 1)))
-        bbox = dummy_draw.textbbox((0, 0), text, font=font)
+        try:
-        text_w = bbox[2] - bbox[0]
+            bbox = dummy_draw.multiline_textbbox((0, 0), text, font=font, spacing=int(font_size * 0.25), align="center")
-        text_h = bbox[3] - bbox[1]
+        except Exception:
+            bbox = dummy_draw.textbbox((0, 0), text, font=font)
+        text_w = (bbox[2] - bbox[0]) if bbox else 0
+        text_h = (bbox[3] - bbox[1]) if bbox else 0
         
-        # 4. 计算总尺寸 (包含 padding, stroke, shadow)
+        # 5. 计算总尺寸 (包含 padding, stroke, shadow)
         strokes = style.get("stroke", [])
         if isinstance(strokes, dict): strokes = [strokes] # 兼容旧格式
         
@@ -156,7 +244,7 @@ class TextRenderer:
         canvas_w = content_w + extra_margin * 2
         canvas_h = content_h + extra_margin * 2
         
-        # 5. 创建画布
+        # 6. 创建画布
         img = Image.new("RGBA", (int(canvas_w), int(canvas_h)), (0, 0, 0, 0))
         draw = ImageDraw.Draw(img)
         
@@ -164,7 +252,7 @@ class TextRenderer:
         center_x = canvas_w // 2
         center_y = canvas_h // 2
         
-        # 6. 绘制顺序: 阴影 -> 背景 -> 描边 -> 文本
+        # 7. 绘制顺序: 阴影 -> 背景 -> 描边 -> 文本
         
         # --- 绘制阴影 (针对整个块) ---
         if shadow:
@@ -183,7 +271,11 @@ class TextRenderer:
                 # 文字阴影
                 txt_x = center_x - text_w / 2
                 txt_y = center_y - text_h / 2
-                shadow_draw.text((txt_x, txt_y), text, font=font, fill=shadow_color)
+                # 多行阴影
+                try:
+                    shadow_draw.multiline_text((txt_x, txt_y), text, font=font, fill=shadow_color, spacing=int(font_size * 0.25), align="center")
+                except Exception:
+                    shadow_draw.text((txt_x, txt_y), text, font=font, fill=shadow_color)
                 # 描边阴影
                 for s in strokes:
                     width = s.get("width", 0)
@@ -217,10 +309,55 @@ class TextRenderer:
             width = s.get("width", 0)
             if width > 0:
                 # 通过偏移模拟描边 (Pillow stroke_width 效果一般，但这里先用原生参数)
-                draw.text((txt_x, txt_y), text, font=font, fill=color, stroke_width=width, stroke_fill=color)
+                try:
+                    draw.multiline_text((txt_x, txt_y), text, font=font, fill=color, spacing=int(font_size * 0.25), align="center", stroke_width=width, stroke_fill=color)
+                except Exception:
+                    draw.text((txt_x, txt_y), text, font=font, fill=color, stroke_width=width, stroke_fill=color)
         
         # --- 绘制文字 ---
-        draw.text((txt_x, txt_y), text, font=font, fill=font_color)
+        # italic：通过仿射变换做简单斜体（先绘制到单独图层，再 shear）
+        # bold：通过多次微小偏移叠加模拟加粗（比改 stroke 更接近“字重”）
+        if italic:
+            text_layer = Image.new("RGBA", img.size, (0, 0, 0, 0))
+            text_draw = ImageDraw.Draw(text_layer)
+            if bold:
+                for dx in (0, 1):
+                    try:
+                        text_draw.multiline_text((txt_x + dx, txt_y), text, font=font, fill=font_color, spacing=int(font_size * 0.25), align="center")
+                    except Exception:
+                        text_draw.text((txt_x + dx, txt_y), text, font=font, fill=font_color)
+            else:
+                try:
+                    text_draw.multiline_text((txt_x, txt_y), text, font=font, fill=font_color, spacing=int(font_size * 0.25), align="center")
+                except Exception:
+                    text_draw.text((txt_x, txt_y), text, font=font, fill=font_color)
+            shear = 0.22  # 经验值：适中倾斜
+            text_layer = text_layer.transform(
+                text_layer.size,
+                Image.AFFINE,
+                (1, shear, 0, 0, 1, 0),
+                resample=Image.BICUBIC
+            )
+            img = Image.alpha_composite(img, text_layer)
+            draw = ImageDraw.Draw(img)
+        else:
+            if bold:
+                for dx in (0, 1):
+                    try:
+                        draw.multiline_text((txt_x + dx, txt_y), text, font=font, fill=font_color, spacing=int(font_size * 0.25), align="center")
+                    except Exception:
+                        draw.text((txt_x + dx, txt_y), text, font=font, fill=font_color)
+            else:
+                try:
+                    draw.multiline_text((txt_x, txt_y), text, font=font, fill=font_color, spacing=int(font_size * 0.25), align="center")
+                except Exception:
+                    draw.text((txt_x, txt_y), text, font=font, fill=font_color)
+
+        # underline：在文本底部画线（与字号相关）
+        if underline:
+            line_y = txt_y + text_h + max(2, int(font_size * 0.08))
+            line_th = max(2, int(font_size * 0.06))
+            draw.rectangle([txt_x, line_y, txt_x + text_w, line_y + line_th], fill=font_color)
         
         # 7. 裁剪多余透明区域
         bbox = img.getbbox()

requirements.txt

@@ -23,6 +23,7 @@ numpy>=1.24.0
 
 # Web UI (Streamlit - 保留原有调试界面)
 streamlit>=1.29.0
+extra-streamlit-components>=0.1.71
 
 # FastAPI Backend (新增前后端分离)
 fastapi>=0.109.0

scripts/import_stickers_manifest.py

@@ -0,0 +1,95 @@
+"""
+根据 manifest 批量导入贴纸到 assets/stickers_builtin，并生成 index.json。
+
+用法：
+  python3 scripts/import_stickers_manifest.py --manifest stickers_manifest.json
+
+manifest 示例：
+{
+  "pack": {
+    "id": "fluent-emoji-subset",
+    "name": "Fluent Emoji 子集（抖音常用）",
+    "license": "MIT (CHECK BEFORE PROD)",
+    "attribution": "Microsoft Fluent UI Emoji"
+  },
+  "categories": [
+    {
+      "id": "douyin-basic",
+      "name": "抖音常用",
+      "items": [
+        {"id": "fire", "name": "火", "url": "https://.../fire.png", "tags": ["火","爆款"]},
+        {"id": "heart", "name": "爱心", "url": "https://.../heart.png", "tags": ["点赞","互动"]}
+      ]
+    }
+  ]
+}
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import os
+from pathlib import Path
+from urllib.request import urlopen, Request
+
+
+def _safe_name(s: str) -> str:
+    return "".join(ch if ch.isalnum() or ch in ("-", "_") else "_" for ch in (s or ""))[:80] or "item"
+
+
+def download(url: str, out: Path) -> None:
+    out.parent.mkdir(parents=True, exist_ok=True)
+    req = Request(url, headers={"User-Agent": "video-flow-stickers/1.0"})
+    with urlopen(req, timeout=60) as r:
+        out.write_bytes(r.read())
+
+
+def main() -> int:
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--manifest", required=True, help="manifest json path")
+    ap.add_argument("--out-dir", default="assets/stickers_builtin", help="output directory")
+    args = ap.parse_args()
+
+    manifest_path = Path(args.manifest)
+    data = json.loads(manifest_path.read_text(encoding="utf-8"))
+
+    out_dir = Path(args.out_dir)
+    out_dir.mkdir(parents=True, exist_ok=True)
+
+    pack = data.get("pack") or {}
+    categories = data.get("categories") or []
+
+    # 下载并重写 file 字段（落地为本地文件）
+    for cat in categories:
+        for it in (cat.get("items") or []):
+            url = str(it.get("url") or "")
+            if not url:
+                continue
+            ext = Path(url.split("?")[0]).suffix.lower()
+            if ext not in [".png", ".svg", ".webp"]:
+                ext = ".png"
+            fid = _safe_name(str(it.get("id") or it.get("name") or "item"))
+            fname = f"{fid}{ext}"
+            target = out_dir / fname
+            if not target.exists():
+                print(f"download: {url} -> {target}")
+                download(url, target)
+            it["file"] = fname
+            it.pop("url", None)
+
+    # 输出 index.json
+    out_index = out_dir / "index.json"
+    out_index.write_text(
+        json.dumps({"pack": pack, "categories": categories}, ensure_ascii=False, indent=2),
+        encoding="utf-8",
+    )
+    print(f"written: {out_index}")
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())
+
+
+

scripts/migrate_projects.py

@@ -0,0 +1,128 @@
+#!/usr/bin/env python3
+"""
+迁移脚本：将旧版 JSON 项目文件导入到 SQLite 数据库（用于 8503 调试）
+
+关键点：
+- 不假设 legacy JSON 与当前 Streamlit UI schema 一致
+- 使用 `modules.legacy_normalizer.normalize_legacy_project()` 做纯规则规范化
+- 保留 `_legacy`，确保信息不丢失
+"""
+import json
+import sys
+from pathlib import Path
+
+# 添加项目根目录到路径
+sys.path.insert(0, str(Path(__file__).parent.parent))
+
+import config
+from modules.db_manager import db
+from modules.legacy_normalizer import normalize_legacy_project
+
+def migrate_json_projects(temp_dir: str = None, force: bool = False):
+    """从 temp 目录读取 project_*.json 文件并导入数据库"""
+    
+    if temp_dir is None:
+        temp_dir = config.TEMP_DIR
+    
+    temp_path = Path(temp_dir)
+    
+    if not temp_path.exists():
+        print(f"❌ temp 目录不存在: {temp_path}")
+        return
+    
+    # 查找所有项目 JSON 文件
+    json_files = list(temp_path.glob("project_*.json"))
+    
+    if not json_files:
+        print(f"⚠️ 未找到项目文件: {temp_path}/project_*.json")
+        return
+    
+    print(f"📂 找到 {len(json_files)} 个项目文件")
+    
+    imported = 0
+    updated = 0
+    skipped = 0
+    errors = 0
+    
+    for json_file in json_files:
+        try:
+            project_id = json_file.stem.replace("project_", "")
+            
+            # 读取 JSON 文件
+            with open(json_file, "r", encoding="utf-8") as f:
+                data = json.load(f)
+            
+            # 检查是否已存在
+            existing = db.get_project(project_id)
+            if existing and not force:
+                print(f"  ⏭️ 跳过已存在: {project_id}")
+                skipped += 1
+                continue
+            
+            # 产品信息：用于 Step1 回显与留存
+            # 注意：legacy 的 image_urls 多为远端 URL；当前 Streamlit Step1 使用 uploaded_images(本地路径)。
+            product_info = {
+                "prompt": data.get("prompt", ""),
+                "image_urls": data.get("image_urls", []),
+                "analysis": data.get("analysis", ""),
+                "questions": data.get("questions", []),
+                "answers": data.get("answers", {}),
+                "uploaded_images": [],  # legacy 无本地上传图路径
+                "_legacy": data,
+            }
+            
+            # 获取项目名称
+            name = data.get("prompt", "")[:50] if data.get("prompt") else f"项目 {project_id}"
+            
+            # 规范化脚本数据：对齐当前 UI schema（并保留 legacy）
+            script_data = normalize_legacy_project(data)
+            
+            if existing and force:
+                # 更新现有项目
+                if script_data:
+                    db.update_project_script(project_id, script_data)
+                status = data.get("status", "draft")
+                db.update_project_status(project_id, status)
+                print(f"  🔄 更新成功: {project_id} ({name[:30]}...)")
+                updated += 1
+            else:
+                # 创建新项目
+                db.create_project(project_id, name, product_info)
+                
+                # 更新脚本
+                if script_data:
+                    db.update_project_script(project_id, script_data)
+                
+                # 更新状态
+                status = data.get("status", "draft")
+                db.update_project_status(project_id, status)
+                
+                print(f"  ✅ 导入成功: {project_id} ({name[:30]}...)")
+                imported += 1
+            
+        except Exception as e:
+            print(f"  ❌ 导入失败 {json_file.name}: {e}")
+            import traceback
+            traceback.print_exc()
+            errors += 1
+    
+    print(f"\n📊 迁移完成:")
+    print(f"   ✅ 新导入: {imported}")
+    print(f"   🔄 已更新: {updated}")
+    print(f"   ⏭️ 已跳过: {skipped}")
+    print(f"   ❌ 失败: {errors}")
+
+
+if __name__ == "__main__":
+    import argparse
+    
+    parser = argparse.ArgumentParser(description="迁移旧版项目到数据库")
+    parser.add_argument("--temp-dir", type=str, default=None, 
+                        help="temp 目录路径 (默认使用 config.TEMP_DIR)")
+    parser.add_argument("--force", action="store_true",
+                        help="强制更新已存在的项目")
+    
+    args = parser.parse_args()
+    
+    print("🚀 开始迁移项目数据...")
+    migrate_json_projects(args.temp_dir, args.force)

scripts/migrate_users_and_owner.py

@@ -0,0 +1,30 @@
+"""
+One-off migration:
+- Ensure admin exists
+- Backfill projects.owner_user_id to admin for legacy projects
+
+Usage:
+  python3 scripts/migrate_users_and_owner.py
+"""
+
+import os
+import sys
+
+# Ensure repo root is on sys.path when executed from scripts/ directory
+REPO_ROOT = os.path.abspath(os.path.join(os.path.dirname(__file__), ".."))
+if REPO_ROOT not in sys.path:
+    sys.path.insert(0, REPO_ROOT)
+
+from modules.db_manager import db
+
+
+def main():
+    admin_id = db.ensure_admin_user("admin", "admin1234")
+    n = db.migrate_projects_owner_to(admin_id)
+    print(f"admin_id={admin_id} backfilled_projects={n}")
+
+
+if __name__ == "__main__":
+    main()
+
+

scripts/scan_legacy_schema.py

@@ -0,0 +1,232 @@
+#!/usr/bin/env python3
+"""
+Scan legacy project JSON schemas under temp dir.
+
+Purpose:
+- Identify schema variants for /opt/gloda-factory/temp/project_*.json
+- Produce a machine-readable summary + a markdown report
+
+This script is READ-ONLY.
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+from collections import Counter, defaultdict
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Any, Dict, List, Tuple
+
+
+def _safe_load_json(path: Path) -> Dict[str, Any] | None:
+    try:
+        return json.loads(path.read_text(encoding="utf-8"))
+    except Exception:
+        return None
+
+
+def _type_name(v: Any) -> str:
+    if v is None:
+        return "null"
+    if isinstance(v, bool):
+        return "bool"
+    if isinstance(v, int):
+        return "int"
+    if isinstance(v, float):
+        return "float"
+    if isinstance(v, str):
+        return "str"
+    if isinstance(v, list):
+        return "list"
+    if isinstance(v, dict):
+        return "dict"
+    return type(v).__name__
+
+
+def _detect_schema_variant(doc: Dict[str, Any]) -> str:
+    """
+    Heuristic:
+    - Schema_A: scenes contain prompt-like fields (image_prompt/visual_prompt/video_prompt)
+    - Schema_B: scenes do NOT contain these, but contain keyframe/story_beat/camera_movement/image_url
+    """
+    scenes = doc.get("scenes") or []
+    if not isinstance(scenes, list):
+        return "Unknown"
+
+    prompt_keys = {"image_prompt", "visual_prompt", "video_prompt"}
+    seen_prompt = False
+    for s in scenes:
+        if isinstance(s, dict) and (set(s.keys()) & prompt_keys):
+            seen_prompt = True
+            break
+
+    if seen_prompt:
+        return "Schema_A"
+
+    # If no prompt keys, but has typical B keys, call Schema_B
+    typical_b = {"keyframe", "story_beat", "camera_movement", "image_url"}
+    seen_b = False
+    for s in scenes:
+        if isinstance(s, dict) and (set(s.keys()) & typical_b):
+            seen_b = True
+            break
+
+    return "Schema_B" if seen_b else "Unknown"
+
+
+@dataclass
+class ScanResult:
+    total_files: int
+    parsed_files: int
+    failed_files: int
+    schema_counts: Counter
+    top_level_key_counts: Counter
+    scene_key_counts: Counter
+    cta_type_counts: Counter
+    sample_by_schema: Dict[str, List[str]]
+
+
+def scan_dir(temp_dir: Path) -> ScanResult:
+    files = sorted(temp_dir.glob("project_*.json"))
+    schema_counts: Counter = Counter()
+    top_level_key_counts: Counter = Counter()
+    scene_key_counts: Counter = Counter()
+    cta_type_counts: Counter = Counter()
+    sample_by_schema: Dict[str, List[str]] = defaultdict(list)
+
+    parsed = 0
+    failed = 0
+
+    for f in files:
+        doc = _safe_load_json(f)
+        if not isinstance(doc, dict):
+            failed += 1
+            continue
+        parsed += 1
+
+        schema = _detect_schema_variant(doc)
+        schema_counts[schema] += 1
+        if len(sample_by_schema[schema]) < 5:
+            pid = str(doc.get("id") or f.stem.replace("project_", ""))
+            sample_by_schema[schema].append(pid)
+
+        # top-level keys
+        for k in doc.keys():
+            top_level_key_counts[k] += 1
+
+        # scenes keys
+        scenes = doc.get("scenes") or []
+        if isinstance(scenes, list):
+            for s in scenes:
+                if isinstance(s, dict):
+                    for k in s.keys():
+                        scene_key_counts[k] += 1
+
+        # cta type
+        cta_type_counts[_type_name(doc.get("cta"))] += 1
+
+    return ScanResult(
+        total_files=len(files),
+        parsed_files=parsed,
+        failed_files=failed,
+        schema_counts=schema_counts,
+        top_level_key_counts=top_level_key_counts,
+        scene_key_counts=scene_key_counts,
+        cta_type_counts=cta_type_counts,
+        sample_by_schema=dict(sample_by_schema),
+    )
+
+
+def _to_jsonable(sr: ScanResult) -> Dict[str, Any]:
+    return {
+        "total_files": sr.total_files,
+        "parsed_files": sr.parsed_files,
+        "failed_files": sr.failed_files,
+        "schema_counts": dict(sr.schema_counts),
+        "cta_type_counts": dict(sr.cta_type_counts),
+        "top_level_key_counts": dict(sr.top_level_key_counts),
+        "scene_key_counts": dict(sr.scene_key_counts),
+        "sample_by_schema": sr.sample_by_schema,
+    }
+
+
+def _render_markdown(sr: ScanResult, temp_dir: Path) -> str:
+    lines: List[str] = []
+    lines.append("# Legacy Project JSON Schema Scan Report\n")
+    lines.append(f"- temp_dir: `{temp_dir}`")
+    lines.append(f"- total_files: {sr.total_files}")
+    lines.append(f"- parsed_files: {sr.parsed_files}")
+    lines.append(f"- failed_files: {sr.failed_files}\n")
+
+    lines.append("## Schema variants\n")
+    for k, v in sr.schema_counts.most_common():
+        samples = ", ".join(sr.sample_by_schema.get(k, [])[:5])
+        lines.append(f"- {k}: {v} (samples: {samples})")
+    lines.append("")
+
+    lines.append("## CTA type distribution\n")
+    for k, v in sr.cta_type_counts.most_common():
+        lines.append(f"- {k}: {v}")
+    lines.append("")
+
+    def _topn(counter: Counter, n: int = 30) -> List[Tuple[str, int]]:
+        return counter.most_common(n)
+
+    lines.append("## Top-level keys (top 30)\n")
+    for k, v in _topn(sr.top_level_key_counts, 30):
+        lines.append(f"- {k}: {v}/{sr.parsed_files}")
+    lines.append("")
+
+    lines.append("## Scene keys (top 40)\n")
+    for k, v in _topn(sr.scene_key_counts, 40):
+        lines.append(f"- {k}: {v}")
+    lines.append("")
+
+    return "\n".join(lines) + "\n"
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser(description="Scan legacy project JSON schemas")
+    parser.add_argument("--temp-dir", required=True, help="Directory containing project_*.json")
+    parser.add_argument("--out-json", required=False, help="Write summary json to path")
+    parser.add_argument("--out-md", required=False, help="Write markdown report to path")
+    args = parser.parse_args()
+
+    temp_dir = Path(args.temp_dir)
+    if not temp_dir.exists():
+        raise SystemExit(f"temp dir not found: {temp_dir}")
+
+    sr = scan_dir(temp_dir)
+    payload = _to_jsonable(sr)
+
+    print(json.dumps(payload, ensure_ascii=False, indent=2))
+
+    if args.out_json:
+        out_json = Path(args.out_json)
+        out_json.parent.mkdir(parents=True, exist_ok=True)
+        out_json.write_text(json.dumps(payload, ensure_ascii=False, indent=2), encoding="utf-8")
+
+    if args.out_md:
+        out_md = Path(args.out_md)
+        out_md.parent.mkdir(parents=True, exist_ok=True)
+        out_md.write_text(_render_markdown(sr, temp_dir), encoding="utf-8")
+
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())
+
+
+
+
+
+
+
+
+
+
+
+
+

web/Dockerfile

@@ -0,0 +1,47 @@
+# Video Flow - React 前端 Dockerfile
+# 多阶段构建
+
+# 构建阶段
+FROM node:20-alpine AS builder
+
+WORKDIR /app
+
+# 复制依赖文件
+COPY package.json package-lock.json* pnpm-lock.yaml* ./
+
+# 安装依赖
+RUN npm install
+
+# 复制源代码
+COPY . .
+
+# 构建
+RUN npm run build
+
+# 生产阶段
+FROM nginx:alpine
+
+# 复制构建产物
+COPY --from=builder /app/dist /usr/share/nginx/html
+
+# 复制 nginx 配置
+COPY nginx.conf /etc/nginx/conf.d/default.conf
+
+# 暴露端口
+EXPOSE 80
+
+CMD ["nginx", "-g", "daemon off;"]
+
+
+
+
+
+
+
+
+
+
+
+
+
+

web/index.html

@@ -0,0 +1,30 @@
+<!DOCTYPE html>
+<html lang="zh-CN">
+  <head>
+    <meta charset="UTF-8" />
+    <link rel="icon" type="image/svg+xml" href="/vite.svg" />
+    <meta name="viewport" content="width=device-width, initial-scale=1.0" />
+    <title>Video Flow Editor</title>
+    <link rel="preconnect" href="https://fonts.googleapis.com">
+    <link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
+    <link href="https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600;700&family=JetBrains+Mono:wght@400;500&display=swap" rel="stylesheet">
+  </head>
+  <body>
+    <div id="root"></div>
+    <script type="module" src="/src/main.tsx"></script>
+  </body>
+</html>
+
+
+
+
+
+
+
+
+
+
+
+
+
+

web/nginx.conf

@@ -0,0 +1,59 @@
+server {
+    listen 80;
+    server_name localhost;
+    root /usr/share/nginx/html;
+    index index.html;
+
+    # Gzip 压缩
+    gzip on;
+    gzip_types text/plain text/css application/json application/javascript text/xml application/xml;
+    gzip_min_length 1000;
+
+    # SPA 路由支持
+    location / {
+        try_files $uri $uri/ /index.html;
+    }
+
+    # API 代理
+    location /api/ {
+        proxy_pass http://api:8000/api/;
+        proxy_http_version 1.1;
+        proxy_set_header Upgrade $http_upgrade;
+        proxy_set_header Connection 'upgrade';
+        proxy_set_header Host $host;
+        proxy_set_header X-Real-IP $remote_addr;
+        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
+        proxy_cache_bypass $http_upgrade;
+        proxy_read_timeout 300;
+        proxy_connect_timeout 300;
+        proxy_send_timeout 300;
+    }
+
+    # 静态资源代理（必须用 ^~ 防止被后面的 regex 缓存规则抢走，导致 /static/* 返回 404）
+    location ^~ /static/ {
+        proxy_pass http://api:8000/static/;
+        proxy_http_version 1.1;
+        proxy_set_header Host $host;
+        proxy_cache_bypass $http_upgrade;
+    }
+
+    # 缓存前端构建产物（仅 /assets/，避免影响 /static/ 代理）
+    location ~* ^/assets/.*\.(js|css|png|jpg|jpeg|gif|ico|svg|woff|woff2)$ {
+        expires 1y;
+        add_header Cache-Control "public, immutable";
+    }
+}
+
+
+
+
+
+
+
+
+
+
+
+
+
+

web/package.json

@@ -0,0 +1,38 @@
+{
+  "name": "video-flow-editor",
+  "private": true,
+  "version": "1.0.0",
+  "type": "module",
+  "scripts": {
+    "dev": "vite",
+    "build": "tsc && vite build",
+    "preview": "vite preview",
+    "lint": "eslint . --ext ts,tsx --report-unused-disable-directives --max-warnings 0"
+  },
+  "dependencies": {
+    "react": "^18.2.0",
+    "react-dom": "^18.2.0",
+    "react-router-dom": "^6.21.0",
+    "@tanstack/react-query": "^5.17.0",
+    "axios": "^1.6.0",
+    "zustand": "^4.4.0",
+    "immer": "^10.0.0",
+    "lucide-react": "^0.303.0",
+    "clsx": "^2.1.0",
+    "tailwind-merge": "^2.2.0",
+    "remotion": "^4.0.0",
+    "@remotion/player": "^4.0.0",
+    "@remotion/cli": "^4.0.0"
+  },
+  "devDependencies": {
+    "@types/react": "^18.2.43",
+    "@types/react-dom": "^18.2.17",
+    "@vitejs/plugin-react": "^4.2.1",
+    "autoprefixer": "^10.4.16",
+    "postcss": "^8.4.32",
+    "tailwindcss": "^3.4.0",
+    "typescript": "^5.3.3",
+    "vite": "^5.0.10"
+  }
+}
+

web/postcss.config.js

@@ -0,0 +1,20 @@
+export default {
+  plugins: {
+    tailwindcss: {},
+    autoprefixer: {},
+  },
+}
+
+
+
+
+
+
+
+
+
+
+
+
+
+

web/src/App.tsx

@@ -0,0 +1,31 @@
+import { BrowserRouter, Routes, Route, Navigate } from 'react-router-dom'
+import { EditorPage } from './pages/EditorPage'
+import { ProjectsPage } from './pages/ProjectsPage'
+
+function App() {
+  return (
+    <BrowserRouter>
+      <Routes>
+        <Route path="/" element={<Navigate to="/projects" replace />} />
+        <Route path="/projects" element={<ProjectsPage />} />
+        <Route path="/editor/:projectId" element={<EditorPage />} />
+      </Routes>
+    </BrowserRouter>
+  )
+}
+
+export default App
+
+
+
+
+
+
+
+
+
+
+
+
+
+

web/src/ClipItem.tsx

@@ -0,0 +1,6 @@
+// Back-compat re-export
+export { ClipItem } from './components/Timeline/ClipItem'
+export { default } from './components/Timeline/ClipItem'
+
+
+

web/src/EditorPage.tsx

@@ -0,0 +1,6 @@
+// Back-compat re-export (some deployments may reference src/EditorPage.tsx)
+export { EditorPage } from './pages/EditorPage'
+export { default } from './pages/EditorPage'
+
+
+

web/src/Timeline.tsx

@@ -0,0 +1,6 @@
+// Back-compat re-export
+export { Timeline } from './components/Timeline/Timeline'
+export { default } from './components/Timeline/Timeline'
+
+
+

web/src/TrackPanel.tsx

@@ -0,0 +1,6 @@
+// Back-compat re-export
+export { TrackPanel } from './components/Timeline/TrackPanel'
+export { default } from './components/Timeline/TrackPanel'
+
+
+

web/src/TrackRow.tsx

@@ -0,0 +1,6 @@
+// Back-compat re-export
+export { TrackRow } from './components/Timeline/TrackRow'
+export { default } from './components/Timeline/TrackRow'
+
+
+

web/src/VideoComposition.tsx

@@ -0,0 +1,6 @@
+// Back-compat re-export
+export { VideoComposition } from './remotion/VideoComposition'
+export { default } from './remotion/VideoComposition'
+
+
+

web/src/components/Timeline/ClipItem.tsx

@@ -0,0 +1,366 @@
+/**
+ * 时间轴片段组件
+ */
+import React, { useRef, useCallback, useEffect, useState } from 'react'
+import { type TimelineClip } from '@/store/editorStore'
+import { cn, timeToPixels, pixelsToTime } from '@/lib/utils'
+
+interface ClipItemProps {
+  clip: TimelineClip
+  trackType: string
+  pixelsPerSecond: number
+  trackHeight: number
+  isSelected: boolean
+  isLocked: boolean
+  onSelect: (e: React.MouseEvent) => void
+  onDrag: (deltaX: number, deltaTime: number) => void
+  onResize: (edge: 'start' | 'end', deltaTime: number) => void
+  onCommit?: () => void
+}
+
+export const ClipItem: React.FC<ClipItemProps> = ({
+  clip,
+  trackType,
+  pixelsPerSecond,
+  trackHeight,
+  isSelected,
+  isLocked,
+  onSelect,
+  onDrag,
+  onResize,
+  onCommit,
+}) => {
+  const clipRef = useRef<HTMLDivElement>(null)
+  const [isDragging, setIsDragging] = useState(false)
+  const [isResizing, setIsResizing] = useState<'start' | 'end' | null>(null)
+  const [thumbUrl, setThumbUrl] = useState<string | null>(null)
+  const [audioPeaks, setAudioPeaks] = useState<number[] | null>(null)
+  
+  const left = timeToPixels(clip.start, pixelsPerSecond)
+  const width = timeToPixels(clip.duration, pixelsPerSecond)
+
+  // 胶片条：对视频片段生成多帧缩略图（MVP），提升辨识度
+  useEffect(() => {
+    if (trackType !== 'video') return
+    if (!clip.sourceUrl) return
+    if (thumbUrl) return
+    // 宽度太小就不抽帧，避免性能压力
+    if (width < 90) return
+    let cancelled = false
+    const src = clip.sourceUrl
+    const minTile = 110
+    const frameCount = Math.max(3, Math.min(8, Math.floor(width / minTile)))
+    const key = `${src}@@${Math.round((clip.trimStart || 0) * 1000)}@@${Math.round((clip.duration || 0) * 1000)}@@${frameCount}`
+    const cache = (globalThis as any).__vfThumbCache as Map<string, string> | undefined
+    if (cache?.has(key)) {
+      setThumbUrl(cache.get(key) || null)
+      return
+    }
+
+    const run = async () => {
+      try {
+        const v = document.createElement('video')
+        v.muted = true
+        v.playsInline = true
+        ;(v as any).crossOrigin = 'anonymous'
+        v.preload = 'metadata'
+        v.src = src
+
+        await new Promise<void>((resolve, reject) => {
+          const onMeta = () => resolve()
+          const onErr = () => reject(new Error('video load error'))
+          v.addEventListener('loadedmetadata', onMeta, { once: true })
+          v.addEventListener('error', onErr, { once: true })
+        })
+
+        const dur = Number.isFinite(v.duration) ? v.duration : 0
+        const t0 = clip.trimStart || 0
+        const clipDur = clip.duration || 0
+        const canvas = document.createElement('canvas')
+        const tileW = 120
+        canvas.width = tileW * frameCount
+        canvas.height = 90
+        const ctx = canvas.getContext('2d')
+        if (!ctx) return
+        ctx.fillStyle = '#000'
+        ctx.fillRect(0, 0, canvas.width, canvas.height)
+
+        for (let i = 0; i < frameCount; i++) {
+          if (cancelled) return
+          const ratio = frameCount === 1 ? 0 : i / (frameCount - 1)
+          const rawT = t0 + ratio * Math.max(0, clipDur - 0.1)
+          const seekT = dur > 0 ? Math.min(Math.max(0, rawT + 0.03), Math.max(0, dur - 0.08)) : Math.max(0, rawT + 0.03)
+          v.currentTime = seekT
+          await new Promise<void>((resolve, reject) => {
+            const onSeek = () => resolve()
+            const onErr = () => reject(new Error('video seek error'))
+            v.addEventListener('seeked', onSeek, { once: true })
+            v.addEventListener('error', onErr, { once: true })
+          })
+          ctx.drawImage(v, i * tileW, 0, tileW, canvas.height)
+          // film-strip separators
+          ctx.fillStyle = 'rgba(0,0,0,0.22)'
+          ctx.fillRect(i * tileW, 0, 1, canvas.height)
+        }
+
+        const dataUrl = canvas.toDataURL('image/jpeg', 0.72)
+        if (cancelled) return
+        setThumbUrl(dataUrl)
+        const c = (globalThis as any).__vfThumbCache || new Map<string, string>()
+        c.set(key, dataUrl)
+        ;(globalThis as any).__vfThumbCache = c
+      } catch {
+        // ignore thumbnail failures
+      }
+    }
+    // 尽量把重活放到空闲时间
+    const ric = (globalThis as any).requestIdleCallback as ((cb: () => void) => number) | undefined
+    ric ? ric(() => run()) : setTimeout(() => run(), 0)
+    return () => { cancelled = true }
+    // eslint-disable-next-line react-hooks/exhaustive-deps
+  }, [trackType, clip.sourceUrl, clip.trimStart, width])
+
+  // 音频波形（MVP）：对 audio/bgm 片段抽样波形（decodeAudioData -> peaks）
+  useEffect(() => {
+    if (trackType !== 'audio' && trackType !== 'bgm') return
+    if (!clip.sourceUrl) return
+    if (audioPeaks) return
+    // 旁白通常片段较短，阈值太大则很难看到波形
+    if (width < 50) return
+    let cancelled = false
+    const src = clip.sourceUrl
+    const points = Math.max(80, Math.min(320, Math.floor(width)))
+    const trimStart = clip.trimStart ?? 0
+    const durReq = clip.duration ?? 0
+    const key = `${src}@@${points}@@${Math.round(trimStart * 1000)}@@${Math.round(durReq * 1000)}`
+    const cache = (globalThis as any).__vfWaveCache as Map<string, number[]> | undefined
+    if (cache?.has(key)) {
+      setAudioPeaks(cache.get(key) || null)
+      return
+    }
+
+    const run = async () => {
+      try {
+        const res = await fetch(src)
+        const buf = await res.arrayBuffer()
+        const ctx = new (window.AudioContext || (window as any).webkitAudioContext)()
+        const audio = await ctx.decodeAudioData(buf.slice(0))
+        const ch = audio.numberOfChannels ? audio.getChannelData(0) : new Float32Array()
+        const sr = audio.sampleRate || 44100
+        const fullLen = ch.length
+        const startS = Math.max(0, trimStart)
+        const endS = Math.max(startS, startS + Math.max(0.01, durReq || audio.duration || 0))
+        const startIdx = Math.min(fullLen, Math.floor(startS * sr))
+        const endIdx = Math.min(fullLen, Math.floor(endS * sr))
+        const len = Math.max(1, endIdx - startIdx)
+        const block = Math.max(1, Math.floor(len / points))
+        const peaks: number[] = new Array(points).fill(0)
+        for (let i = 0; i < points; i++) {
+          const start = startIdx + i * block
+          const end = Math.min(endIdx, start + block)
+          let max = 0
+          for (let j = start; j < end; j++) {
+            const v = Math.abs(ch[j])
+            if (v > max) max = v
+          }
+          peaks[i] = max
+        }
+        // normalize
+        const m = Math.max(...peaks, 1e-6)
+        const norm = peaks.map(p => p / m)
+        if (cancelled) return
+        setAudioPeaks(norm)
+        const c = (globalThis as any).__vfWaveCache || new Map<string, number[]>()
+        c.set(key, norm)
+        ;(globalThis as any).__vfWaveCache = c
+      } catch {
+        // ignore
+      }
+    }
+    const ric = (globalThis as any).requestIdleCallback as ((cb: () => void) => number) | undefined
+    ric ? ric(() => run()) : setTimeout(() => run(), 0)
+    return () => { cancelled = true }
+    // eslint-disable-next-line react-hooks/exhaustive-deps
+  }, [trackType, clip.sourceUrl, width])
+  
+  // 获取轨道颜色
+  const getTrackColor = () => {
+    switch (trackType) {
+      case 'video': return 'bg-track-video'
+      case 'audio': return 'bg-track-voiceover'
+      case 'subtitle': return 'bg-track-subtitle'
+      case 'fancy_text': return 'bg-track-fancy'
+      case 'bgm': return 'bg-track-bgm'
+      default: return 'bg-gray-500'
+    }
+  }
+  
+  // 处理拖拽开始
+  const handleMouseDown = useCallback((e: React.MouseEvent) => {
+    if (isResizing) return
+    e.stopPropagation()
+
+    // 锁定轨道：仍允许选中查看属性，但禁止拖拽移动
+    onSelect(e)
+    if (isLocked) return
+
+    setIsDragging(true)
+    
+    const startX = e.clientX
+    
+    const handleMouseMove = (moveEvent: MouseEvent) => {
+      const deltaX = moveEvent.clientX - startX
+      const deltaTime = pixelsToTime(deltaX, pixelsPerSecond)
+      onDrag(deltaX, deltaTime)
+    }
+    
+    const handleMouseUp = () => {
+      setIsDragging(false)
+      document.removeEventListener('mousemove', handleMouseMove)
+      document.removeEventListener('mouseup', handleMouseUp)
+      onCommit?.()
+    }
+    
+    document.addEventListener('mousemove', handleMouseMove)
+    document.addEventListener('mouseup', handleMouseUp)
+  }, [clip.start, pixelsPerSecond, isLocked, isResizing, onSelect, onDrag])
+  
+  // 处理调整大小开始
+  const handleResizeStart = useCallback((
+    e: React.MouseEvent,
+    edge: 'start' | 'end'
+  ) => {
+    if (isLocked) return
+    e.stopPropagation()
+    
+    setIsResizing(edge)
+    const startX = e.clientX
+    
+    const handleMouseMove = (moveEvent: MouseEvent) => {
+      const deltaX = moveEvent.clientX - startX
+      const deltaTime = pixelsToTime(edge === 'start' ? deltaX : deltaX, pixelsPerSecond)
+      onResize(edge, deltaTime)
+    }
+    
+    const handleMouseUp = () => {
+      setIsResizing(null)
+      document.removeEventListener('mousemove', handleMouseMove)
+      document.removeEventListener('mouseup', handleMouseUp)
+      onCommit?.()
+    }
+    
+    document.addEventListener('mousemove', handleMouseMove)
+    document.addEventListener('mouseup', handleMouseUp)
+  }, [pixelsPerSecond, isLocked, onResize])
+  
+  // 获取显示内容
+  const getClipContent = () => {
+    if (clip.text) {
+      return clip.text.slice(0, 20) + (clip.text.length > 20 ? '...' : '')
+    }
+    if (clip.sourceUrl) {
+      return (clip.sourceUrl || '').split('/').pop()?.slice(0, 15) || 'Video'
+    }
+    return clip.type
+  }
+
+  const transitionBadge = (() => {
+    if (trackType !== 'video') return null
+    const s: any = clip.style || {}
+    const t = String(s.vTransitionType ?? s.v_transition_type ?? '')
+    if (!t) return null
+    const map: Record<string, string> = {
+      fade: '淡出',
+      fadeWhite: '淡到白',
+      flash: '闪白',
+      zoomOut: '缩小',
+      zoomIn: '推进',
+      slideLeft: '左滑',
+      slideRight: '右滑',
+      slideUp: '上滑',
+      slideDown: '下滑',
+      rotateOut: '旋转',
+      blurOut: '模糊',
+      blurFade: '模糊淡出',
+      desaturate: '去饱和',
+      colorPop: '色彩增强',
+      hueShift: '色相偏移',
+      darken: '变暗',
+    }
+    const name = map[t] || t
+    return (
+      <div className="absolute left-1 top-1 z-10 px-1.5 py-0.5 rounded bg-black/45 text-[10px] text-white/90 pointer-events-none">
+        {name}
+      </div>
+    )
+  })()
+  
+  return (
+    <div
+      ref={clipRef}
+      className={cn(
+        'vf-clip',
+        'absolute rounded overflow-hidden cursor-grab select-none',
+        'transition-shadow',
+        getTrackColor(),
+        isSelected && 'ring-2 ring-white ring-offset-1 ring-offset-editor-bg',
+        isDragging && 'cursor-grabbing opacity-90 shadow-lg',
+        isLocked && 'cursor-not-allowed'
+      )}
+      style={{
+        left,
+        width: Math.max(width, 20),
+        top: 4,
+        height: trackHeight - 8,
+        backgroundImage: thumbUrl ? `linear-gradient(rgba(0,0,0,0.35), rgba(0,0,0,0.35)), url(${thumbUrl})` : undefined,
+        backgroundSize: thumbUrl ? 'auto 100%' : undefined,
+        backgroundRepeat: thumbUrl ? 'repeat-x' : undefined,
+        backgroundPosition: 'center',
+      }}
+      onMouseDown={handleMouseDown}
+    >
+      {transitionBadge}
+
+      {/* 左边调整手柄 */}
+      <div
+        className={cn(
+          'absolute left-0 top-0 bottom-0 w-2 cursor-ew-resize',
+          'hover:bg-white/20'
+        )}
+        onMouseDown={(e) => handleResizeStart(e, 'start')}
+      />
+      
+      {/* 内容 */}
+      <div className="absolute inset-0 px-2 flex items-center overflow-hidden pointer-events-none">
+        <span className="text-xs text-white/90 font-medium truncate">
+          {getClipContent()}
+        </span>
+      </div>
+
+      {/* 音频波形（MVP） */}
+      {(trackType === 'audio' || trackType === 'bgm') && audioPeaks && (
+        <div className="absolute left-0 right-0 bottom-1 h-6 px-1 opacity-90 pointer-events-none">
+          <svg width="100%" height="100%" viewBox={`0 0 ${audioPeaks.length} 20`} preserveAspectRatio="none">
+            {audioPeaks.map((p, i) => {
+              const h = Math.max(1, Math.round(p * 18))
+              const y = 10 - h / 2
+              return <rect key={i} x={i} y={y} width={0.9} height={h} fill="rgba(255,255,255,0.72)" />
+            })}
+          </svg>
+        </div>
+      )}
+      
+      {/* 右边调整手柄 */}
+      <div
+        className={cn(
+          'absolute right-0 top-0 bottom-0 w-2 cursor-ew-resize',
+          'hover:bg-white/20'
+        )}
+        onMouseDown={(e) => handleResizeStart(e, 'end')}
+      />
+    </div>
+  )
+}
+
+export default ClipItem
+

web/src/components/Timeline/Playhead.tsx

@@ -0,0 +1,44 @@
+/**
+ * 播放头组件
+ */
+import React from 'react'
+import { timeToPixels } from '@/lib/utils'
+
+interface PlayheadProps {
+  currentTime: number
+  pixelsPerSecond: number
+  height: number
+  onPointerDown?: (e: React.PointerEvent<HTMLDivElement>) => void
+}
+
+export const Playhead: React.FC<PlayheadProps> = ({
+  currentTime,
+  pixelsPerSecond,
+  height,
+  onPointerDown,
+}) => {
+  const left = timeToPixels(currentTime, pixelsPerSecond)
+  
+  return (
+    <div className="playhead" style={{ left, height }}>
+      {/* 增大命中区域，提升“拖拽跟手” */}
+      <div className="absolute -left-2 top-0 bottom-0 w-4 cursor-ew-resize" onPointerDown={onPointerDown} />
+    </div>
+  )
+}
+
+export default Playhead
+
+
+
+
+
+
+
+
+
+
+
+
+
+

web/src/components/Timeline/TimeRuler.tsx

@@ -0,0 +1,72 @@
+/**
+ * 时间标尺组件
+ */
+import React, { useMemo } from 'react'
+import { formatTimeShort } from '@/lib/utils'
+
+interface TimeRulerProps {
+  duration: number
+  pixelsPerSecond: number
+  scrollX?: number
+}
+
+export const TimeRuler: React.FC<TimeRulerProps> = ({
+  duration,
+  pixelsPerSecond,
+}) => {
+  // 计算刻度间隔
+  const { majorInterval, minorInterval } = useMemo(() => {
+    if (pixelsPerSecond >= 100) {
+      return { majorInterval: 1, minorInterval: 0.5 }
+    } else if (pixelsPerSecond >= 50) {
+      return { majorInterval: 2, minorInterval: 1 }
+    } else if (pixelsPerSecond >= 25) {
+      return { majorInterval: 5, minorInterval: 1 }
+    } else {
+      return { majorInterval: 10, minorInterval: 5 }
+    }
+  }, [pixelsPerSecond])
+  
+  // 生成刻度
+  const ticks = useMemo(() => {
+    const result: { time: number; isMajor: boolean }[] = []
+    
+    for (let t = 0; t <= duration; t += minorInterval) {
+      const isMajor = t % majorInterval === 0
+      result.push({ time: t, isMajor })
+    }
+    
+    return result
+  }, [duration, majorInterval, minorInterval])
+  
+  return (
+    <div className="relative h-full bg-editor-panel">
+      {ticks.map(({ time, isMajor }) => {
+        const x = time * pixelsPerSecond
+        
+        return (
+          <div
+            key={time}
+            className="absolute top-0 flex flex-col items-center"
+            style={{ left: x }}
+          >
+            {/* 刻度线 */}
+            <div
+              className={isMajor ? 'w-px h-3 bg-editor-text-muted' : 'w-px h-2 bg-editor-border'}
+            />
+            
+            {/* 时间标签 */}
+            {isMajor && (
+              <span className="text-[10px] text-editor-text-muted mt-0.5">
+                {formatTimeShort(time)}
+              </span>
+            )}
+          </div>
+        )
+      })}
+    </div>
+  )
+}
+
+export default TimeRuler
+

web/src/components/Timeline/Timeline.tsx

@@ -0,0 +1,408 @@
+/**
+ * 时间轴编辑器组件
+ */
+import React, { useEffect, useMemo, useRef, useCallback, useState } from 'react'
+import { useEditorStore } from '@/store/editorStore'
+import { TrackRow } from './TrackRow'
+import { TimeRuler } from './TimeRuler'
+import { Playhead } from './Playhead'
+import { cn, timeToPixels, pixelsToTime } from '@/lib/utils'
+import { Scissors, Trash2 } from 'lucide-react'
+
+const PIXELS_PER_SECOND_BASE = 50
+const TRACK_HEIGHT = 48
+const RULER_HEIGHT = 28
+
+export const Timeline: React.FC = () => {
+  const containerRef = useRef<HTMLDivElement>(null)
+  const scrollContainerRef = useRef<HTMLDivElement>(null)
+  const contentRef = useRef<HTMLDivElement>(null)
+  
+  const {
+    tracks,
+    totalDuration,
+    currentTime,
+    zoom,
+    scrollX,
+    snapGuideTime,
+    setSelectedClips,
+    clearSelection,
+    setCurrentTime,
+    setScrollX,
+    setZoom,
+    pushHistory,
+    splitAtTime,
+    deleteAtPlayhead,
+    addClip,
+  } = useEditorStore()
+
+  const visibleTracks = useMemo(() => tracks.filter(t => !t.collapsed), [tracks])
+  
+  const [isDraggingPlayhead, setIsDraggingPlayhead] = useState(false)
+  const [viewportWidth, setViewportWidth] = useState(800)
+  const [boxSel, setBoxSel] = useState<null | { x1: number; y1: number; x2: number; y2: number }>(null)
+  const dragRafRef = useRef<number | null>(null)
+  
+  const pixelsPerSecond = PIXELS_PER_SECOND_BASE * zoom
+  const timelineWidth = Math.max(totalDuration * pixelsPerSecond, 800)
+
+  // 可视窗口（用于虚拟化渲染，减少大量 clip 节点导致的重排/卡顿）
+  const viewWindow = useMemo(() => {
+    const startPx = scrollX
+    const endPx = scrollX + viewportWidth
+    const bufferPx = Math.max(200, viewportWidth * 0.5)
+    const viewStartTime = pixelsToTime(Math.max(0, startPx - bufferPx), pixelsPerSecond)
+    const viewEndTime = pixelsToTime(endPx + bufferPx, pixelsPerSecond)
+    return { viewStartTime, viewEndTime }
+  }, [scrollX, viewportWidth, pixelsPerSecond])
+
+  // 监听容器宽度变化（ResizeObserver）
+  useEffect(() => {
+    const el = scrollContainerRef.current
+    if (!el) return
+    const update = () => setViewportWidth(el.clientWidth || 800)
+    update()
+    const ro = new ResizeObserver(update)
+    ro.observe(el)
+    return () => ro.disconnect()
+  }, [])
+  
+  // 处理时间轴点击 - 移动播放头
+  const handleTimelineClick = useCallback((e: React.MouseEvent) => {
+    if (isDraggingPlayhead) return
+    
+    const rect = e.currentTarget.getBoundingClientRect()
+    const x = e.clientX - rect.left + scrollX
+    const time = pixelsToTime(x, pixelsPerSecond)
+    setCurrentTime(Math.max(0, Math.min(time, totalDuration)))
+  }, [pixelsPerSecond, scrollX, totalDuration, isDraggingPlayhead])
+  
+  const updatePlayheadFromClientX = useCallback((clientX: number) => {
+    const scrollEl = scrollContainerRef.current
+    const rulerEl = containerRef.current
+    if (!scrollEl || !rulerEl) return
+    // 以“标尺区域的可视窗口”为基准计算 x，并叠加实时 scrollLeft（不要用滞后的 store.scrollX）
+    const rulerRect = rulerEl.getBoundingClientRect()
+    const x = clientX - rulerRect.left - 128 /* track label width */ + (scrollEl.scrollLeft || 0)
+    const time = pixelsToTime(x, pixelsPerSecond)
+    const next = Math.max(0, Math.min(time, totalDuration))
+    // raf 合并高频更新，避免拖拽卡顿
+    if (dragRafRef.current) cancelAnimationFrame(dragRafRef.current)
+    dragRafRef.current = requestAnimationFrame(() => {
+      setCurrentTime(next)
+      dragRafRef.current = null
+    })
+  }, [pixelsPerSecond, totalDuration, setCurrentTime])
+
+  // 处理播放头拖拽（Pointer capture：跟手、不会丢事件）
+  const handlePlayheadPointerDown = useCallback((e: React.PointerEvent) => {
+    e.preventDefault()
+    e.stopPropagation()
+    setIsDraggingPlayhead(true)
+    ;(e.currentTarget as HTMLElement).setPointerCapture?.(e.pointerId)
+    updatePlayheadFromClientX(e.clientX)
+  }, [updatePlayheadFromClientX])
+
+  const handlePlayheadPointerMove = useCallback((e: React.PointerEvent) => {
+    if (!isDraggingPlayhead) return
+    e.preventDefault()
+    updatePlayheadFromClientX(e.clientX)
+  }, [isDraggingPlayhead, updatePlayheadFromClientX])
+
+  const handlePlayheadPointerUp = useCallback((e: React.PointerEvent) => {
+    if (!isDraggingPlayhead) return
+    e.preventDefault()
+    setIsDraggingPlayhead(false)
+    try { (e.currentTarget as HTMLElement).releasePointerCapture?.(e.pointerId) } catch {}
+  }, [isDraggingPlayhead])
+  
+  // 滚轮缩放
+  const handleWheel = useCallback((e: React.WheelEvent) => {
+    if (e.ctrlKey || e.metaKey) {
+      e.preventDefault()
+      const delta = e.deltaY > 0 ? 0.9 : 1.1
+      setZoom(zoom * delta)
+    } else {
+      // 水平滚动
+      setScrollX(scrollX + e.deltaX)
+    }
+  }, [zoom, scrollX])
+  
+  // 同步滚动
+  const handleScroll = useCallback((e: React.UIEvent<HTMLDivElement>) => {
+    setScrollX(e.currentTarget.scrollLeft)
+  }, [])
+
+  // 框选（只在空白区域拖拽触发）
+  const handleContentMouseDown = useCallback((e: React.MouseEvent) => {
+    if (e.button !== 0) return
+    if (isDraggingPlayhead) return
+    const target = e.target as HTMLElement
+    if (target?.closest?.('.vf-clip')) return
+    if (!scrollContainerRef.current || !contentRef.current) return
+
+    const scrollEl = scrollContainerRef.current
+    const rect = contentRef.current.getBoundingClientRect()
+    const startX = e.clientX - rect.left + scrollEl.scrollLeft
+    const startY = e.clientY - rect.top + scrollEl.scrollTop
+    setBoxSel({ x1: startX, y1: startY, x2: startX, y2: startY })
+
+    const onMove = (me: MouseEvent) => {
+      const x = me.clientX - rect.left + scrollEl.scrollLeft
+      const y = me.clientY - rect.top + scrollEl.scrollTop
+      setBoxSel((cur) => cur ? ({ ...cur, x2: x, y2: y }) : null)
+    }
+    const onUp = () => {
+      document.removeEventListener('mousemove', onMove)
+      document.removeEventListener('mouseup', onUp)
+      setBoxSel((cur) => {
+        if (!cur) return null
+        const xMin = Math.min(cur.x1, cur.x2)
+        const xMax = Math.max(cur.x1, cur.x2)
+        const yMin = Math.min(cur.y1, cur.y2)
+        const yMax = Math.max(cur.y1, cur.y2)
+        const timeMin = pixelsToTime(xMin, pixelsPerSecond)
+        const timeMax = pixelsToTime(xMax, pixelsPerSecond)
+
+        // 小拖拽视为“点空白取消选择”
+        if (Math.abs(xMax - xMin) < 4 && Math.abs(yMax - yMin) < 4) {
+          clearSelection()
+          return null
+        }
+
+        const picked: string[] = []
+        visibleTracks.forEach((t, idx) => {
+          const top = idx * TRACK_HEIGHT
+          const bottom = top + TRACK_HEIGHT
+          if (bottom < yMin || top > yMax) return
+          for (const c of t.clips) {
+            const start = c.start ?? 0
+            const end = start + (c.duration ?? 0)
+            if (end < timeMin || start > timeMax) continue
+            picked.push(c.id)
+          }
+        })
+        setSelectedClips(picked)
+        return null
+      })
+    }
+    document.addEventListener('mousemove', onMove)
+    document.addEventListener('mouseup', onUp)
+  }, [visibleTracks, pixelsPerSecond, isDraggingPlayhead, clearSelection, setSelectedClips])
+
+  const handleDropAsset = useCallback((e: React.DragEvent) => {
+    e.preventDefault()
+    e.stopPropagation()
+    const raw = e.dataTransfer.getData('application/json') || ''
+    if (!raw) return
+    let payload: any = null
+    try { payload = JSON.parse(raw) } catch { payload = null }
+    if (!payload) return
+
+    const vTrack = tracks.find(t => t.type === 'video')
+    const stickerTrack = tracks.find(t => t.type === 'sticker')
+
+    const scrollEl = scrollContainerRef.current
+    const rect = contentRef.current?.getBoundingClientRect()
+    const x = rect ? (e.clientX - rect.left + (scrollEl?.scrollLeft || 0)) : (scrollEl?.scrollLeft || 0)
+    const t = Math.max(0, pixelsToTime(x, pixelsPerSecond))
+
+    if (payload.kind === 'asset' && payload.assetType === 'video') {
+      if (!vTrack) return
+      const url = String(payload.url || '')
+      if (!url) return
+      pushHistory({ label: '添加素材', icon: 'media' })
+      addClip(vTrack.id, {
+        type: 'video',
+        start: t,
+        duration: 3,
+        trimStart: 0,
+        trimEnd: 3,
+        sourceUrl: url,
+        sourcePath: payload.localPath ? String(payload.localPath) : undefined,
+      })
+      return
+    }
+
+    if (payload.kind === 'sticker') {
+      if (!stickerTrack) return
+      const url = String(payload.url || '')
+      if (!url) return
+      pushHistory({ label: '添加贴纸', icon: 'sticker' })
+      addClip(stickerTrack.id, {
+        type: 'sticker',
+        start: t,
+        duration: 2,
+        sourceUrl: url,
+        text: String(payload.name || ''),
+        position: { x: 0.8, y: 0.2 },
+        style: { scale: 1.0, rotate: 0 },
+      })
+      return
+    }
+  }, [tracks, pixelsPerSecond, pushHistory, addClip])
+  
+  return (
+    <div
+      ref={containerRef}
+      className="h-full flex flex-col bg-editor-bg select-none"
+      onWheel={handleWheel}
+    >
+      {/* 时间标尺 */}
+      <div
+        className="shrink-0 bg-editor-panel border-b border-editor-border overflow-hidden"
+        style={{ height: RULER_HEIGHT }}
+      >
+        <div className="flex">
+          {/* 轨道标签占位 */}
+          <div className="w-32 shrink-0 bg-editor-panel" />
+          
+          {/* 标尺 */}
+          <div
+            className="relative cursor-pointer"
+            style={{ width: timelineWidth }}
+            onClick={handleTimelineClick}
+            onPointerMove={handlePlayheadPointerMove}
+            onPointerUp={handlePlayheadPointerUp}
+            onPointerCancel={handlePlayheadPointerUp}
+          >
+            <TimeRuler
+              duration={totalDuration}
+              pixelsPerSecond={pixelsPerSecond}
+              scrollX={scrollX}
+            />
+            
+            {/* 播放头顶部标记 */}
+            <div
+              className="absolute top-0 w-4 h-4 -ml-2 cursor-ew-resize"
+              style={{ left: timeToPixels(currentTime, pixelsPerSecond) - scrollX }}
+              onPointerDown={handlePlayheadPointerDown}
+            >
+              <div className="w-0 h-0 border-l-[6px] border-l-transparent border-r-[6px] border-r-transparent border-t-[8px] border-t-red-500" />
+            </div>
+
+            {/* 红线附近快捷操作：剪切/删除（不跟着素材 hover 走） */}
+            <div
+              className="absolute top-1 -ml-2 flex gap-1 z-20"
+              style={{ left: timeToPixels(currentTime, pixelsPerSecond) - scrollX }}
+              onPointerDown={(e) => e.stopPropagation()}
+            >
+              <button
+                type="button"
+                className="w-7 h-7 rounded bg-black/55 hover:bg-black/70 flex items-center justify-center"
+                title="剪切（以红线为准）"
+                onClick={() => {
+                  pushHistory({ label: '剪切', icon: 'scissors' })
+                  splitAtTime(currentTime)
+                }}
+              >
+                <Scissors className="w-4 h-4 text-white/90" />
+              </button>
+              <button
+                type="button"
+                className="w-7 h-7 rounded bg-black/55 hover:bg-black/70 flex items-center justify-center"
+                title="删除（红线所在片段）"
+                onClick={() => {
+                  pushHistory({ label: '删除片段', icon: 'trash' })
+                  deleteAtPlayhead()
+                }}
+              >
+                <Trash2 className="w-4 h-4 text-white/90" />
+              </button>
+            </div>
+          </div>
+        </div>
+      </div>
+      
+      {/* 轨道区域 */}
+      <div
+        ref={scrollContainerRef}
+        className="flex-1 overflow-auto"
+        onScroll={handleScroll}
+      >
+        <div className="flex min-h-full">
+          {/* 轨道标签 */}
+          <div className="w-32 shrink-0 bg-editor-panel border-r border-editor-border sticky left-0 z-10">
+            {visibleTracks.map(track => (
+              <div
+                key={track.id}
+                className={cn(
+                  'flex items-center px-3 border-b border-editor-border',
+                  'text-sm text-editor-text-muted'
+                )}
+                style={{ height: TRACK_HEIGHT }}
+              >
+                <div
+                  className={cn(
+                    'w-1.5 h-4 rounded-sm mr-2',
+                    track.type === 'video' && 'bg-track-video',
+                    track.type === 'audio' && 'bg-track-voiceover',
+                    track.type === 'subtitle' && 'bg-track-subtitle',
+                    track.type === 'fancy_text' && 'bg-track-fancy',
+                    track.type === 'bgm' && 'bg-track-bgm',
+                  )}
+                />
+                <span className="truncate">{track.name}</span>
+              </div>
+            ))}
+          </div>
+          
+          {/* 轨道内容 */}
+          <div
+            className="relative"
+            style={{ width: timelineWidth, minHeight: visibleTracks.length * TRACK_HEIGHT }}
+            ref={contentRef}
+            onMouseDown={handleContentMouseDown}
+            onDragOver={(e) => { e.preventDefault(); e.dataTransfer.dropEffect = 'copy' }}
+            onDrop={handleDropAsset}
+          >
+            {visibleTracks.map((track, index) => (
+              <TrackRow
+                key={track.id}
+                track={track}
+                pixelsPerSecond={pixelsPerSecond}
+                trackHeight={TRACK_HEIGHT}
+                top={index * TRACK_HEIGHT}
+                viewStartTime={viewWindow.viewStartTime}
+                viewEndTime={viewWindow.viewEndTime}
+                onClipChange={() => pushHistory()}
+              />
+            ))}
+            
+            {/* 播放头线 */}
+            <Playhead
+              currentTime={currentTime}
+              pixelsPerSecond={pixelsPerSecond}
+              height={visibleTracks.length * TRACK_HEIGHT}
+              onPointerDown={handlePlayheadPointerDown}
+            />
+
+            {/* 吸附/对齐辅助线 */}
+            {typeof snapGuideTime === 'number' && (
+              <div
+                className="absolute top-0 bottom-0 w-px bg-yellow-400/80 pointer-events-none"
+                style={{ left: timeToPixels(snapGuideTime, pixelsPerSecond) }}
+              />
+            )}
+
+            {/* 框选矩形 */}
+            {boxSel && (
+              <div
+                className="absolute border border-editor-accent bg-editor-accent/10 pointer-events-none"
+                style={{
+                  left: Math.min(boxSel.x1, boxSel.x2),
+                  top: Math.min(boxSel.y1, boxSel.y2),
+                  width: Math.abs(boxSel.x2 - boxSel.x1),
+                  height: Math.abs(boxSel.y2 - boxSel.y1),
+                }}
+              />
+            )}
+          </div>
+        </div>
+      </div>
+    </div>
+  )
+}
+
+export default Timeline
+

web/src/components/Timeline/TrackPanel.tsx

@@ -0,0 +1,872 @@
+/**
+ * 轨道属性面板
+ */
+import React, { useEffect, useState } from 'react'
+import { useMutation, useQuery } from '@tanstack/react-query'
+import {
+  Volume2,
+  VolumeX,
+  Lock,
+  Unlock,
+  Trash2,
+  RefreshCw,
+  Headphones,
+  ChevronRight,
+  ChevronDown,
+  ChevronUp,
+  AlertTriangle,
+} from 'lucide-react'
+import { useEditorStore } from '@/store/editorStore'
+import { editorApi, assetsApi } from '@/lib/api'
+import { cn, formatTimeShort, generateId, parseTimeInput } from '@/lib/utils'
+
+export const TrackPanel: React.FC = () => {
+  const {
+    tracks,
+    selectedClipId,
+    selectedClipIds,
+    soloTrackIds,
+    toggleTrackSolo,
+    toggleTrackMute,
+    toggleTrackLock,
+    toggleTrackCollapse,
+    updateClip,
+    deleteClip,
+    groupSelected,
+    ungroupSelected,
+    pushHistory,
+  } = useEditorStore()
+
+  const { data: fontList } = useQuery({
+    queryKey: ['font-list'],
+    queryFn: assetsApi.getFonts,
+  })
+  
+  // 找到选中的片段
+  const selectedClip = tracks
+    .flatMap(t => t.clips.map(c => ({ ...c, trackId: t.id, trackType: t.type })))
+    .find(c => c.id === selectedClipId)
+  
+  // TTS 重新生成
+  const ttsMutation = useMutation({
+    mutationFn: (params: { text: string; targetDuration?: number }) =>
+      editorApi.generateVoiceover(params.text, undefined, params.targetDuration),
+    onSuccess: (data) => {
+      if (selectedClip && data.url) {
+        updateClip(selectedClip.trackId, selectedClip.id, {
+          sourceUrl: data.url,
+          sourcePath: data.path,
+          sourceDuration: typeof data.duration === 'number' && Number.isFinite(data.duration) ? data.duration : (selectedClip as any).sourceDuration,
+          needsVoiceoverRegenerate: false,
+        })
+        pushHistory()
+      }
+    },
+  })
+  
+  // 花字重新生成
+  const fancyTextMutation = useMutation({
+    mutationFn: (params: { text: string; style?: Record<string, unknown> }) =>
+      editorApi.generateFancyText(params.text, params.style),
+    onSuccess: (data) => {
+      if (selectedClip && data.url) {
+        updateClip(selectedClip.trackId, selectedClip.id, {
+          sourceUrl: data.url,
+          sourcePath: data.path,
+        })
+        pushHistory()
+      }
+    },
+  })
+  
+  const [editingText, setEditingText] = useState('')
+  const [isEditing, setIsEditing] = useState(false)
+  const [showAdvanced, setShowAdvanced] = useState(false)
+
+  // 基础：更友好的时间输入（M:SS）
+  const [basicTime, setBasicTime] = useState({ start: '', end: '' })
+  // 高级：精确输入（秒）
+  const [timing, setTiming] = useState({ start: '', duration: '', trimStart: '', trimEnd: '' })
+
+  useEffect(() => {
+    if (!selectedClip) return
+    const s = Number(selectedClip.start ?? 0)
+    const d = Number(selectedClip.duration ?? 0)
+    setBasicTime({
+      start: formatTimeShort(Math.max(0, s)),
+      end: formatTimeShort(Math.max(0, s + d)),
+    })
+    setTiming({
+      start: String(selectedClip.start ?? 0),
+      duration: String(selectedClip.duration ?? 0),
+      trimStart: String(selectedClip.trimStart ?? 0),
+      trimEnd: String(selectedClip.trimEnd ?? ((selectedClip.trimStart ?? 0) + (selectedClip.duration ?? 0))),
+    })
+  }, [selectedClipId, selectedClip?.start, selectedClip?.duration, selectedClip?.trimStart, selectedClip?.trimEnd])
+  
+  // 开始编辑文本
+  const startEditing = () => {
+    if (selectedClip?.text) {
+      setEditingText(selectedClip.text)
+      setIsEditing(true)
+    }
+  }
+  
+  // 保存文本
+  const saveText = () => {
+    if (!selectedClip) return
+    
+    updateClip(selectedClip.trackId, selectedClip.id, { text: editingText })
+    setIsEditing(false)
+    pushHistory()
+  }
+
+  const saveStyle = (patch: Record<string, unknown>) => {
+    if (!selectedClip) return
+    updateClip(selectedClip.trackId, selectedClip.id, { style: { ...(selectedClip.style || {}), ...patch } })
+    pushHistory()
+  }
+
+  const saveTiming = () => {
+    if (!selectedClip) return
+    const toNum = (v: string) => {
+      const n = Number(v)
+      return Number.isFinite(n) ? n : 0
+    }
+    const minDur = 0.5
+
+    let start = Math.max(0, toNum(timing.start))
+    let duration = Math.max(minDur, toNum(timing.duration))
+    let trimStart = Math.max(0, toNum(timing.trimStart))
+    let trimEnd = toNum(timing.trimEnd)
+    if (!Number.isFinite(trimEnd) || trimEnd <= 0) trimEnd = trimStart + duration
+
+    // normalize trimEnd to duration
+    trimEnd = trimStart + duration
+
+    updateClip(selectedClip.trackId, selectedClip.id, { start, duration, trimStart, trimEnd })
+    pushHistory()
+  }
+
+  const saveBasicTime = () => {
+    if (!selectedClip) return
+    const minDur = 0.5
+    const start = parseTimeInput(basicTime.start)
+    const end = parseTimeInput(basicTime.end)
+    if (start == null || end == null) return
+    const s = Math.max(0, start)
+    const d = Math.max(minDur, end - s)
+    updateClip(selectedClip.trackId, selectedClip.id, { start: s, duration: d })
+    pushHistory()
+  }
+
+  const saveAudioParams = (params: { volume?: number; fadeIn?: number; fadeOut?: number; ducking?: boolean; duckVolume?: number; playbackRate?: number }) => {
+    if (!selectedClip) return
+    updateClip(selectedClip.trackId, selectedClip.id, {
+      ...(typeof params.volume === 'number' && Number.isFinite(params.volume) ? { volume: Math.max(0, Math.min(1, params.volume)) } : {}),
+      ...(typeof params.fadeIn === 'number' && Number.isFinite(params.fadeIn) ? { fadeIn: Math.max(0, params.fadeIn) } : {}),
+      ...(typeof params.fadeOut === 'number' && Number.isFinite(params.fadeOut) ? { fadeOut: Math.max(0, params.fadeOut) } : {}),
+      ...(typeof params.ducking === 'boolean' ? { ducking: params.ducking } : {}),
+      ...(typeof params.duckVolume === 'number' && Number.isFinite(params.duckVolume) ? { duckVolume: Math.max(0.05, Math.min(1, params.duckVolume)) } : {}),
+      ...(typeof params.playbackRate === 'number' && Number.isFinite(params.playbackRate) ? { playbackRate: Math.max(0.5, Math.min(2.0, params.playbackRate)) } : {}),
+    })
+    pushHistory()
+  }
+  
+  // 重新生成 TTS
+  const regenerateTTS = () => {
+    if (!selectedClip?.text) return
+    ttsMutation.mutate({
+      text: selectedClip.text,
+      targetDuration: selectedClip.duration,
+    })
+  }
+  
+  // 重新生成花字
+  const regenerateFancyText = () => {
+    if (!selectedClip?.text) return
+    fancyTextMutation.mutate({
+      text: selectedClip.text,
+      style: selectedClip.style,
+    })
+  }
+  
+  // 删除片段
+  const handleDeleteClip = () => {
+    if (!selectedClip) return
+    deleteClip(selectedClip.trackId, selectedClip.id)
+    pushHistory()
+  }
+
+  // 绑定到视频片段（用于字幕/旁白联动）
+  const bindToVideo = () => {
+    if (!selectedClip) return
+    // 只对字幕/旁白有意义
+    const isSubtitle = selectedClip.trackType === 'subtitle'
+    const isVoice = selectedClip.trackId === 'audio-voiceover'
+    if (!isSubtitle && !isVoice) return
+
+    const videoTrack = tracks.find(t => t.type === 'video')
+    if (!videoTrack) return
+    const s = selectedClip.start ?? 0
+    const e = s + (selectedClip.duration ?? 0)
+
+    // 选 overlap 最大的视频片段
+    let best: any = null
+    let bestOverlap = 0
+    for (const v of videoTrack.clips) {
+      const vs = v.start ?? 0
+      const ve = vs + (v.duration ?? 0)
+      const overlap = Math.max(0, Math.min(e, ve) - Math.max(s, vs))
+      if (overlap > bestOverlap) {
+        bestOverlap = overlap
+        best = v
+      }
+    }
+    if (!best) return
+
+    const gid = best.groupId || `grp_${generateId()}`
+    if (!best.groupId) {
+      updateClip(videoTrack.id, best.id, { groupId: gid })
+    }
+    updateClip(selectedClip.trackId, selectedClip.id, { groupId: gid })
+    pushHistory()
+  }
+  
+  return (
+    <div className="h-full flex flex-col">
+      {/* 轨道列表 */}
+      <div className="p-4 border-b border-editor-border">
+        <h3 className="text-sm font-medium text-editor-text mb-2">轨道</h3>
+        <p className="text-xs text-editor-text-muted mb-3">
+          剪切：移动红线到想切的位置，点“剪切”或按 <span className="font-mono">S</span>。拖动片段会自动把后面的片段推开。
+        </p>
+        <div className="space-y-2">
+          {tracks.map(track => (
+            <div
+              key={track.id}
+              className={cn(
+                'flex items-center justify-between p-2 rounded-lg',
+                'bg-editor-surface hover:bg-editor-hover transition-colors'
+              )}
+            >
+              <div className="flex items-center gap-2 min-w-0">
+                <button
+                  onClick={() => toggleTrackCollapse(track.id)}
+                  className="p-1 rounded text-editor-text-muted hover:text-editor-text"
+                  title={track.collapsed ? '展开轨道' : '折叠轨道'}
+                >
+                  {track.collapsed ? <ChevronRight className="w-4 h-4" /> : <ChevronDown className="w-4 h-4" />}
+                </button>
+                <span className="text-sm text-editor-text truncate">{track.name}</span>
+              </div>
+              <div className="flex items-center gap-1">
+                {showAdvanced && (
+                  <button
+                    onClick={() => toggleTrackSolo(track.id)}
+                    className={cn(
+                      'p-1 rounded',
+                      (soloTrackIds || []).includes(track.id) ? 'text-green-400' : 'text-editor-text-muted hover:text-editor-text'
+                    )}
+                    title="只听这一条轨道（高级）"
+                  >
+                    <Headphones className="w-4 h-4" />
+                  </button>
+                )}
+                <button
+                  onClick={() => toggleTrackMute(track.id)}
+                  className={cn(
+                    'p-1 rounded',
+                    track.muted ? 'text-red-400' : 'text-editor-text-muted hover:text-editor-text'
+                  )}
+                  title={track.muted ? '取消静音' : '静音'}
+                >
+                  {track.muted ? <VolumeX className="w-4 h-4" /> : <Volume2 className="w-4 h-4" />}
+                </button>
+                <button
+                  onClick={() => toggleTrackLock(track.id)}
+                  className={cn(
+                    'p-1 rounded',
+                    track.locked ? 'text-yellow-400' : 'text-editor-text-muted hover:text-editor-text'
+                  )}
+                  title={track.locked ? '已锁定：不可拖动/修改' : '锁定：防误触'}
+                >
+                  {track.locked ? <Lock className="w-4 h-4" /> : <Unlock className="w-4 h-4" />}
+                </button>
+              </div>
+            </div>
+          ))}
+        </div>
+        <button
+          onClick={() => setShowAdvanced(v => !v)}
+          className="mt-3 w-full py-1.5 rounded bg-editor-surface text-editor-text text-sm hover:bg-editor-hover flex items-center justify-center gap-1"
+        >
+          {showAdvanced ? <ChevronUp className="w-4 h-4" /> : <ChevronDown className="w-4 h-4" />}
+          {showAdvanced ? '收起高级选项' : '显示高级选项'}
+        </button>
+      </div>
+      
+      {/* 选中片段属性 */}
+      {selectedClip && (
+        <div className="p-4 border-b border-editor-border">
+          <h3 className="text-sm font-medium text-editor-text mb-3">片段</h3>
+          
+          <div className="space-y-3">
+            {/* 基础时间（新手友好） */}
+            <div className="space-y-2">
+              <div className="text-xs text-editor-text-muted">时间（M:SS）</div>
+              <div className="grid grid-cols-2 gap-2">
+                <div>
+                  <label className="text-editor-text-muted text-xs">开始</label>
+                  <input
+                    value={basicTime.start}
+                    onChange={(e) => setBasicTime({ ...basicTime, start: e.target.value })}
+                    className={cn('w-full px-2 py-1 rounded text-sm font-mono', 'bg-editor-bg border border-editor-border text-editor-text')}
+                    placeholder="0:00"
+                  />
+                </div>
+                <div>
+                  <label className="text-editor-text-muted text-xs">结束</label>
+                  <input
+                    value={basicTime.end}
+                    onChange={(e) => setBasicTime({ ...basicTime, end: e.target.value })}
+                    className={cn('w-full px-2 py-1 rounded text-sm font-mono', 'bg-editor-bg border border-editor-border text-editor-text')}
+                    placeholder="0:02"
+                  />
+                </div>
+              </div>
+              <button
+                onClick={saveBasicTime}
+                className="w-full py-1.5 rounded bg-editor-surface text-editor-text text-sm hover:bg-editor-hover"
+              >
+                应用
+              </button>
+              <div className="text-xs text-editor-text-muted">
+                提示：更常用的是直接在时间轴拖动片段和边缘。
+              </div>
+            </div>
+
+            {/* 高级：精确时间/裁剪（秒） */}
+            {showAdvanced && (
+              <div className="space-y-2 pt-2 border-t border-editor-border">
+                <div className="text-xs text-editor-text-muted">精确调整（秒）</div>
+                <div className="grid grid-cols-2 gap-2">
+                  <div>
+                    <label className="text-editor-text-muted text-xs">开始(s)</label>
+                    <input
+                      value={timing.start}
+                      onChange={(e) => setTiming({ ...timing, start: e.target.value })}
+                      className={cn('w-full px-2 py-1 rounded text-sm font-mono', 'bg-editor-bg border border-editor-border text-editor-text')}
+                    />
+                  </div>
+                  <div>
+                    <label className="text-editor-text-muted text-xs">时长(s)</label>
+                    <input
+                      value={timing.duration}
+                      onChange={(e) => setTiming({ ...timing, duration: e.target.value })}
+                      className={cn('w-full px-2 py-1 rounded text-sm font-mono', 'bg-editor-bg border border-editor-border text-editor-text')}
+                    />
+                  </div>
+                </div>
+                {selectedClip.trackType === 'video' && (
+                  <div className="grid grid-cols-2 gap-2">
+                    <div>
+                      <label className="text-editor-text-muted text-xs">trimStart(s)</label>
+                      <input
+                        value={timing.trimStart}
+                        onChange={(e) => setTiming({ ...timing, trimStart: e.target.value })}
+                        className={cn('w-full px-2 py-1 rounded text-sm font-mono', 'bg-editor-bg border border-editor-border text-editor-text')}
+                      />
+                    </div>
+                    <div>
+                      <label className="text-editor-text-muted text-xs">trimEnd(s)</label>
+                      <input
+                        value={timing.trimEnd}
+                        onChange={(e) => setTiming({ ...timing, trimEnd: e.target.value })}
+                        className={cn('w-full px-2 py-1 rounded text-sm font-mono', 'bg-editor-bg border border-editor-border text-editor-text')}
+                      />
+                    </div>
+                  </div>
+                )}
+                <button
+                  onClick={saveTiming}
+                  className="w-full py-1.5 rounded bg-editor-surface text-editor-text text-sm hover:bg-editor-hover"
+                >
+                  应用精确参数
+                </button>
+              </div>
+            )}
+            
+            {/* 文本编辑 */}
+            {selectedClip.text !== undefined && (
+              <div>
+                <label className="text-editor-text-muted text-xs block mb-1">文本</label>
+                {(selectedClip.trackType === 'subtitle' || selectedClip.trackType === 'fancy_text') && !isEditing && (
+                  <div className="mb-2 text-[11px] text-editor-text-muted">
+                    建议：在画面上<strong>双击文字</strong>直接编辑（更像抖音/剪映）。这里只保留为“高级兜底”。
+                  </div>
+                )}
+                {isEditing ? (
+                  <div className="space-y-2">
+                    <textarea
+                      value={editingText}
+                      onChange={(e) => setEditingText(e.target.value)}
+                      className={cn(
+                        'w-full px-3 py-2 rounded-lg text-sm',
+                        'bg-editor-bg border border-editor-border',
+                        'text-editor-text focus:outline-none focus:border-editor-accent'
+                      )}
+                      rows={3}
+                    />
+                    <div className="flex gap-2">
+                      <button
+                        onClick={saveText}
+                        className="flex-1 py-1.5 rounded bg-editor-accent text-white text-sm font-medium"
+                      >
+                        保存
+                      </button>
+                      <button
+                        onClick={() => setIsEditing(false)}
+                        className="flex-1 py-1.5 rounded bg-editor-surface text-editor-text text-sm"
+                      >
+                        取消
+                      </button>
+                    </div>
+                  </div>
+                ) : (
+                  <div className="space-y-2">
+                    <div className={cn('px-3 py-2 rounded-lg text-sm', 'bg-editor-bg border border-editor-border', 'text-editor-text')}>
+                      {selectedClip.text || '（空）'}
+                    </div>
+                    <button
+                      onClick={startEditing}
+                      className="w-full py-1.5 rounded bg-editor-surface text-editor-text text-sm hover:bg-editor-hover"
+                    >
+                      在右侧编辑（高级）
+                    </button>
+                  </div>
+                )}
+              </div>
+            )}
+
+            {/* 字幕/花字/贴纸 */}
+            {(selectedClip.trackType === 'subtitle' || selectedClip.trackType === 'fancy_text' || selectedClip.trackType === 'sticker') && (
+              <div className="space-y-2 pt-2 border-t border-editor-border">
+                <div className="text-xs text-editor-text-muted">
+                  {selectedClip.trackType === 'sticker' ? '贴纸（位置 / 大小）' : '样式（字体 / 颜色 / B I U）'}
+                </div>
+
+                {selectedClip.trackType === 'sticker' && (
+                  <div className="space-y-2">
+                    <div className="grid grid-cols-2 gap-2">
+                      <div>
+                        <label className="text-editor-text-muted text-xs block mb-1">大小</label>
+                        <input
+                          type="range"
+                          min={50}
+                          max={200}
+                          step={5}
+                          value={Math.round((Number((selectedClip.style as any)?.scale ?? 1) || 1) * 100)}
+                          onChange={(e) => saveStyle({ scale: Math.max(0.5, Math.min(2.0, Number(e.target.value) / 100)) })}
+                          className="w-full"
+                        />
+                        <div className="text-xs text-editor-text-muted text-right">
+                          {(Number((selectedClip.style as any)?.scale ?? 1) || 1).toFixed(2)}x
+                        </div>
+                      </div>
+                      <div>
+                        <label className="text-editor-text-muted text-xs block mb-1">旋转</label>
+                        <input
+                          type="range"
+                          min={-180}
+                          max={180}
+                          step={5}
+                          value={Math.round(Number((selectedClip.style as any)?.rotate ?? 0) || 0)}
+                          onChange={(e) => saveStyle({ rotate: Number(e.target.value) || 0 })}
+                          className="w-full"
+                        />
+                        <div className="text-xs text-editor-text-muted text-right">
+                          {Math.round(Number((selectedClip.style as any)?.rotate ?? 0) || 0)}°
+                        </div>
+                      </div>
+                    </div>
+                    <div className="grid grid-cols-2 gap-2">
+                      <div>
+                        <label className="text-editor-text-muted text-xs block mb-1">X（0~1）</label>
+                        <input
+                          type="number"
+                          step={0.01}
+                          value={String(Number((selectedClip.position as any)?.x ?? 0.8))}
+                          onChange={(e) => {
+                            const base = (selectedClip.position || { x: 0.8, y: 0.2 }) as any
+                            updateClip(selectedClip.trackId, selectedClip.id, { position: { x: Math.max(0, Math.min(1, Number(e.target.value) || 0)), y: base.y ?? 0.2 } })
+                          }}
+                          className={cn('w-full px-2 py-1 rounded text-sm font-mono', 'bg-editor-bg border border-editor-border text-editor-text')}
+                        />
+                      </div>
+                      <div>
+                        <label className="text-editor-text-muted text-xs block mb-1">Y（0~1）</label>
+                        <input
+                          type="number"
+                          step={0.01}
+                          value={String(Number((selectedClip.position as any)?.y ?? 0.2))}
+                          onChange={(e) => {
+                            const base = (selectedClip.position || { x: 0.8, y: 0.2 }) as any
+                            updateClip(selectedClip.trackId, selectedClip.id, { position: { x: base.x ?? 0.8, y: Math.max(0, Math.min(1, Number(e.target.value) || 0)) } })
+                          }}
+                          className={cn('w-full px-2 py-1 rounded text-sm font-mono', 'bg-editor-bg border border-editor-border text-editor-text')}
+                        />
+                      </div>
+                    </div>
+                    <div className="text-[11px] text-editor-text-muted">
+                      提示：后续会支持在画面中直接拖拽贴纸定位（和字幕一致）。
+                    </div>
+                  </div>
+                )}
+
+                {selectedClip.trackType !== 'sticker' && (
+                <>
+                <div className="grid grid-cols-2 gap-2">
+                  <div className="col-span-2">
+                    <label className="text-editor-text-muted text-xs block mb-1">字体</label>
+                    <select
+                      value={String(((selectedClip.style as any)?.font_family ?? '') as any)}
+                      onChange={(e) => {
+                        const v = e.target.value
+                        saveStyle({ font_family: v || undefined })
+                      }}
+                      className={cn('w-full px-2 py-1 rounded text-sm', 'bg-editor-bg border border-editor-border text-editor-text')}
+                    >
+                      <option value="">默认</option>
+                      {(Array.isArray(fontList) ? fontList : []).map((f: any) => {
+                        // custom font：用 filename 作为 font_family（renderer 可解析，预览也能 FontFace 加载）
+                        // system font：用绝对 path 便于导出；预览侧会自动回退为 PingFang
+                        const val = f.url ? (f.css_family || f.id) : (f.path || f.id)
+                        return (
+                          <option key={String(f.id)} value={String(val)}>
+                            {String(f.name || f.id)}
+                          </option>
+                        )
+                      })}
+                    </select>
+                  </div>
+                  <div>
+                    <label className="text-editor-text-muted text-xs block mb-1">字号</label>
+                    <input
+                      type="number"
+                      min={10}
+                      max={160}
+                      step={1}
+                      value={String(Number((selectedClip.style as any)?.font_size ?? (selectedClip.trackType === 'subtitle' ? 60 : 72)))}
+                      onChange={(e) => saveStyle({ font_size: Math.max(10, Math.min(160, Number(e.target.value) || 60)) })}
+                      className={cn('w-full px-2 py-1 rounded text-sm font-mono', 'bg-editor-bg border border-editor-border text-editor-text')}
+                    />
+                  </div>
+                  <div>
+                    <label className="text-editor-text-muted text-xs block mb-1">颜色</label>
+                    <input
+                      type="color"
+                      value={String((selectedClip.style as any)?.font_color ?? '#ffffff')}
+                      onChange={(e) => saveStyle({ font_color: e.target.value })}
+                      className="w-full h-9 rounded bg-editor-bg border border-editor-border"
+                    />
+                  </div>
+                  <div className="col-span-2">
+                    <label className="text-editor-text-muted text-xs block mb-1">文本框宽度（控制换行）</label>
+                    <input
+                      type="range"
+                      min={20}
+                      max={100}
+                      step={1}
+                      value={Math.round((Number((selectedClip.style as any)?.box_w ?? (selectedClip.trackType === 'subtitle' ? 0.8 : 0.7)) || 0.8) * 100)}
+                      onChange={(e) => saveStyle({ box_w: Math.max(0.2, Math.min(1, Number(e.target.value) / 100)) })}
+                      className="w-full"
+                    />
+                    <div className="text-xs text-editor-text-muted text-right">
+                      {Math.round((Number((selectedClip.style as any)?.box_w ?? (selectedClip.trackType === 'subtitle' ? 0.8 : 0.7)) || 0.8) * 100)}%
+                    </div>
+                    <div className="text-[11px] text-editor-text-muted">
+                      也可以直接在画面上选中文字，拖右侧手柄调整宽度。
+                    </div>
+                  </div>
+                </div>
+
+                <div className="flex items-center gap-2">
+                  <button
+                    className={cn('px-2 py-1 rounded text-sm', (selectedClip.style as any)?.bold ? 'bg-editor-accent text-white' : 'bg-editor-surface text-editor-text')}
+                    onClick={() => saveStyle({ bold: !((selectedClip.style as any)?.bold) })}
+                    title="加粗"
+                  >
+                    B
+                  </button>
+                  <button
+                    className={cn('px-2 py-1 rounded text-sm italic', (selectedClip.style as any)?.italic ? 'bg-editor-accent text-white' : 'bg-editor-surface text-editor-text')}
+                    onClick={() => saveStyle({ italic: !((selectedClip.style as any)?.italic) })}
+                    title="斜体"
+                  >
+                    I
+                  </button>
+                  <button
+                    className={cn('px-2 py-1 rounded text-sm underline', (selectedClip.style as any)?.underline ? 'bg-editor-accent text-white' : 'bg-editor-surface text-editor-text')}
+                    onClick={() => saveStyle({ underline: !((selectedClip.style as any)?.underline) })}
+                    title="下划线"
+                  >
+                    U
+                  </button>
+                  <div className="flex-1" />
+                  <div className="text-xs text-editor-text-muted">提示：选中后可在预览里拖动位置</div>
+                </div>
+                </>
+                )}
+              </div>
+            )}
+
+            {/* 视频转场（基础：淡入/淡出到黑，不需要重叠） */}
+            {showAdvanced && selectedClip.trackType === 'video' && (
+              <div className="space-y-2 pt-2 border-t border-editor-border">
+                <div className="text-xs text-editor-text-muted">转场（基础）</div>
+                <div className="grid grid-cols-2 gap-2">
+                  <div>
+                    <label className="text-editor-text-muted text-xs">淡入(s)</label>
+                    <input
+                      type="number"
+                      step="0.05"
+                      value={String(Number((selectedClip.style as any)?.vFadeIn ?? (selectedClip.style as any)?.v_fade_in ?? 0))}
+                      onChange={(e) => saveStyle({ vFadeIn: Math.max(0, Number(e.target.value) || 0) })}
+                      className={cn('w-full px-2 py-1 rounded text-sm font-mono', 'bg-editor-bg border border-editor-border text-editor-text')}
+                    />
+                  </div>
+                  <div>
+                    <label className="text-editor-text-muted text-xs">淡出(s)</label>
+                    <input
+                      type="number"
+                      step="0.05"
+                      value={String(Number((selectedClip.style as any)?.vFadeOut ?? (selectedClip.style as any)?.v_fade_out ?? 0))}
+                      onChange={(e) => saveStyle({ vFadeOut: Math.max(0, Number(e.target.value) || 0) })}
+                      className={cn('w-full px-2 py-1 rounded text-sm font-mono', 'bg-editor-bg border border-editor-border text-editor-text')}
+                    />
+                  </div>
+                </div>
+                <div className="text-xs text-editor-text-muted">
+                  说明：这是“淡入/淡出到黑”的基础转场（不需要片段重叠）。更高级的“交叉溶解”以后再加。
+                </div>
+              </div>
+            )}
+
+            {/* 音频参数（旁白/BGM/视频原声） */}
+            {(selectedClip.trackType === 'audio' || selectedClip.trackType === 'bgm' || selectedClip.trackType === 'video') && (
+              <div className="space-y-2">
+                <div className="text-xs text-editor-text-muted">声音</div>
+                {selectedClip.trackId === 'audio-voiceover' && (
+                  <div>
+                    <label className="text-editor-text-muted text-xs">旁白倍速（纯播放）</label>
+                    <div className="flex items-center gap-2">
+                      <input
+                        type="range"
+                        min={50}
+                        max={200}
+                        step={5}
+                        value={Math.round((Number((selectedClip as any).playbackRate ?? 1) || 1) * 100)}
+                        onChange={(e) => saveAudioParams({ playbackRate: Math.max(0.5, Math.min(2.0, Number(e.target.value) / 100)) })}
+                        className="flex-1"
+                      />
+                      <span className="text-xs text-editor-text-muted w-12 text-right">
+                        {(Number((selectedClip as any).playbackRate ?? 1) || 1).toFixed(2)}x
+                      </span>
+                    </div>
+                    <div className="text-[11px] text-editor-text-muted">
+                      提示：拖拽旁白片段时长会自动计算倍速；也可手动调倍速（预览/导出一致）。
+                    </div>
+                  </div>
+                )}
+                <div>
+                  <label className="text-editor-text-muted text-xs">音量</label>
+                  <div className="flex items-center gap-2">
+                    <input
+                      type="range"
+                      min={0}
+                      max={100}
+                      value={Math.round((((selectedClip as any).volume ?? (selectedClip.trackType === 'bgm' ? 0.15 : selectedClip.trackType === 'video' ? 0.2 : 1)) as number) * 100)}
+                      onChange={(e) => saveAudioParams({ volume: Number(e.target.value) / 100 })}
+                      className="flex-1"
+                    />
+                    <span className="text-xs text-editor-text-muted w-10 text-right">
+                      {Math.round((((selectedClip as any).volume ?? (selectedClip.trackType === 'bgm' ? 0.15 : selectedClip.trackType === 'video' ? 0.2 : 1)) as number) * 100)}%
+                    </span>
+                  </div>
+                </div>
+                {showAdvanced && (
+                  <div className="grid grid-cols-2 gap-2 pt-1">
+                    <div>
+                      <label className="text-editor-text-muted text-xs">淡入(s)</label>
+                      <input
+                        type="number"
+                        step="0.05"
+                        value={String((selectedClip as any).fadeIn ?? (selectedClip.trackType === 'bgm' ? 0.8 : 0.05))}
+                        onChange={(e) => saveAudioParams({ fadeIn: Number(e.target.value) })}
+                        className={cn('w-full px-2 py-1 rounded text-sm font-mono', 'bg-editor-bg border border-editor-border text-editor-text')}
+                      />
+                    </div>
+                    <div>
+                      <label className="text-editor-text-muted text-xs">淡出(s)</label>
+                      <input
+                        type="number"
+                        step="0.05"
+                        value={String((selectedClip as any).fadeOut ?? (selectedClip.trackType === 'bgm' ? 0.8 : 0.05))}
+                        onChange={(e) => saveAudioParams({ fadeOut: Number(e.target.value) })}
+                        className={cn('w-full px-2 py-1 rounded text-sm font-mono', 'bg-editor-bg border border-editor-border text-editor-text')}
+                      />
+                    </div>
+                  </div>
+                )}
+
+                {/* BGM 自动压低（高级） */}
+                {showAdvanced && selectedClip.trackType === 'bgm' && (
+                  <div className="space-y-2 pt-1">
+                    <label className="flex items-center justify-between text-sm text-editor-text">
+                      <span>自动压低背景音乐</span>
+                      <input
+                        type="checkbox"
+                        checked={typeof (selectedClip as any).ducking === 'boolean' ? (selectedClip as any).ducking : true}
+                        onChange={(e) => saveAudioParams({ ducking: e.target.checked })}
+                      />
+                    </label>
+                    <div>
+                      <label className="text-editor-text-muted text-xs">压低到</label>
+                      <input
+                        type="range"
+                        min={5}
+                        max={100}
+                        value={Math.round((((selectedClip as any).duckVolume ?? 0.25) as number) * 100)}
+                        onChange={(e) => saveAudioParams({ duckVolume: Number(e.target.value) / 100 })}
+                        className="w-full"
+                      />
+                      <div className="text-xs text-editor-text-muted text-right">
+                        {Math.round((((selectedClip as any).duckVolume ?? 0.25) as number) * 100)}%
+                      </div>
+                    </div>
+                  </div>
+                )}
+              </div>
+            )}
+            
+            {/* 操作按钮 */}
+            <div className="flex gap-2">
+              {selectedClip.trackId === 'audio-voiceover' && (
+                <button
+                  onClick={regenerateTTS}
+                  disabled={ttsMutation.isPending}
+                  className={cn(
+                    'flex-1 flex items-center justify-center gap-1.5 py-2 rounded-lg text-sm',
+                    'bg-editor-surface text-editor-text hover:bg-editor-hover transition-colors',
+                    ttsMutation.isPending && 'opacity-50'
+                  )}
+                >
+                  <RefreshCw className={cn('w-4 h-4', ttsMutation.isPending && 'animate-spin')} />
+                  {(selectedClip as any).needsVoiceoverRegenerate ? '重新生成（适配时长）' : '生成配音'}
+                </button>
+              )}
+              
+              {selectedClip.trackType === 'fancy_text' && (
+                <button
+                  onClick={regenerateFancyText}
+                  disabled={fancyTextMutation.isPending}
+                  className={cn(
+                    'flex-1 flex items-center justify-center gap-1.5 py-2 rounded-lg text-sm',
+                    'bg-editor-surface text-editor-text hover:bg-editor-hover transition-colors',
+                    fancyTextMutation.isPending && 'opacity-50'
+                  )}
+                >
+                  <RefreshCw className={cn('w-4 h-4', fancyTextMutation.isPending && 'animate-spin')} />
+                  重新生成
+                </button>
+              )}
+              
+              <button
+                onClick={handleDeleteClip}
+                className={cn(
+                  'p-2 rounded-lg text-red-400 hover:bg-red-500/10 transition-colors'
+                )}
+              >
+                <Trash2 className="w-4 h-4" />
+              </button>
+            </div>
+
+            {/* 分组/联动 */}
+            {showAdvanced && (
+              <div className="pt-2 border-t border-editor-border">
+              <div className="flex items-center justify-between">
+                <span className="text-xs text-editor-text-muted">组</span>
+                <span className="text-xs text-editor-text">
+                  {(selectedClip as any).groupId ? String((selectedClip as any).groupId) : '—'}
+                </span>
+              </div>
+              <div className="flex gap-2 mt-2">
+                <button
+                  onClick={() => { groupSelected(); pushHistory() }}
+                  disabled={(selectedClipIds || []).length < 2}
+                  className={cn(
+                    'flex-1 py-1.5 rounded text-sm',
+                    'bg-editor-surface text-editor-text hover:bg-editor-hover',
+                    (selectedClipIds || []).length < 2 && 'opacity-50'
+                  )}
+                >
+                  分组（多选）
+                </button>
+                <button
+                  onClick={() => { ungroupSelected(); pushHistory() }}
+                  className={cn('flex-1 py-1.5 rounded text-sm', 'bg-editor-surface text-editor-text hover:bg-editor-hover')}
+                >
+                  取消分组
+                </button>
+              </div>
+              <button
+                onClick={bindToVideo}
+                className={cn(
+                  'w-full mt-2 py-1.5 rounded text-sm',
+                  'bg-editor-surface text-editor-text hover:bg-editor-hover'
+                )}
+              >
+                绑定到视频片段（字幕/旁白跟随）
+              </button>
+              <p className="text-xs text-editor-text-muted mt-1">
+                绑定后拖动视频片段会带着同组字幕/旁白一起移动。
+              </p>
+              </div>
+            )}
+          </div>
+        </div>
+      )}
+
+      {/* 空状态 */}
+      {!selectedClip && (
+        <div className="p-4">
+          <h3 className="text-sm font-medium text-editor-text mb-2">片段属性</h3>
+          <p className="text-sm text-editor-text-muted">
+            点击时间轴上的片段后，这里会出现“文本 / 音量 / 时间”等常用设置。
+          </p>
+        </div>
+      )}
+
+      {/* 旁白时长/文本变更提示（新手避免误解“拖拉=变速”） */}
+      {selectedClip && selectedClip.trackId === 'audio-voiceover' && (selectedClip as any).needsVoiceoverRegenerate && (
+        <div className="p-4 border-t border-editor-border">
+          <div className="p-3 rounded-lg bg-yellow-500/10 border border-yellow-500/30">
+            <div className="flex items-start gap-2">
+              <AlertTriangle className="w-4 h-4 text-yellow-400 mt-0.5" />
+              <div className="min-w-0">
+                <div className="text-sm text-editor-text font-medium">旁白时长已改变</div>
+                <div className="text-xs text-editor-text-muted mt-1">
+                  提示：拖动/拉伸旁白片段不会自动改变语速。要让旁白“变慢/变快并贴合当前时长”，请点击上方“重新生成（适配时长）”。
+                </div>
+              </div>
+            </div>
+          </div>
+        </div>
+      )}
+    </div>
+  )
+}
+
+export default TrackPanel
+

web/src/components/Timeline/TrackRow.tsx

@@ -0,0 +1,171 @@
+/**
+ * 轨道行组件
+ */
+import React, { useCallback } from 'react'
+import { useEditorStore, type Track } from '@/store/editorStore'
+import { ClipItem } from './ClipItem'
+import { cn } from '@/lib/utils'
+import { applyDragRules, applyResizeEndRules, applyResizeStartRules } from '@/lib/timelineRules'
+
+interface TrackRowProps {
+  track: Track
+  pixelsPerSecond: number
+  trackHeight: number
+  top: number
+  viewStartTime?: number
+  viewEndTime?: number
+  onClipChange?: () => void
+}
+
+export const TrackRow: React.FC<TrackRowProps> = ({
+  track,
+  pixelsPerSecond,
+  trackHeight,
+  top,
+  viewStartTime,
+  viewEndTime,
+  onClipChange,
+}) => {
+  const {
+    selectedClipIds,
+    toggleClipSelection,
+    updateClip,
+    currentTime,
+    rippleMode,
+    rippleShiftFrom,
+    setSnapGuideTime,
+    shiftGroup,
+  } = useEditorStore()
+  
+  // 处理片段拖拽
+  const handleClipDrag = useCallback((
+    clipId: string,
+    _deltaX: number,
+    deltaTime: number
+  ) => {
+    const clip = track.clips.find(c => c.id === clipId)
+    if (!clip || track.locked) return
+    
+    const desiredStart = Math.max(0, clip.start + deltaTime)
+    // 分组联动：同 groupId 的片段跨轨一起移动（不参与 ripple 规则）
+    if (clip.groupId) {
+      const nextStart = applyDragRules(track.clips, clipId, desiredStart, currentTime)
+      const delta = nextStart - clip.start
+      if (Math.abs(delta) > 1e-6) {
+        setSnapGuideTime(nextStart)
+        shiftGroup(clip.groupId, delta)
+      }
+      return
+    }
+    if (track.type === 'video' && rippleMode) {
+      // ripple: move this clip and all following clips together (preserve spacing)
+      const ordered = track.clips.slice().sort((a, b) => a.start - b.start)
+      const idx = ordered.findIndex(c => c.id === clipId)
+      if (idx === -1) return
+      const prev = idx > 0 ? ordered[idx - 1] : null
+      const minStart = prev ? (prev.start + prev.duration) : 0
+      const nextStart = Math.max(minStart, desiredStart)
+      const delta = nextStart - clip.start
+      if (Math.abs(delta) > 1e-6) {
+        rippleShiftFrom(track.id, clipId, delta)
+      }
+      return
+    }
+
+    const nextStart = applyDragRules(track.clips, clipId, desiredStart, currentTime)
+    // 若发生吸附/修正，显示辅助线
+    if (Math.abs(nextStart - desiredStart) > 1e-4) setSnapGuideTime(nextStart)
+    updateClip(track.id, clipId, { start: nextStart })
+  }, [track, updateClip, currentTime, rippleMode, rippleShiftFrom, setSnapGuideTime, shiftGroup])
+  
+  // 处理片段调整大小
+  const handleClipResize = useCallback((
+    clipId: string,
+    edge: 'start' | 'end',
+    deltaTime: number
+  ) => {
+    const clip = track.clips.find(c => c.id === clipId)
+    if (!clip || track.locked) return
+    
+    if (edge === 'start') {
+      const desiredStart = clip.start + deltaTime
+      const next = applyResizeStartRules(track.clips, clipId, desiredStart, currentTime)
+      updateClip(track.id, clipId, next)
+    } else {
+      const desiredEnd = clip.start + clip.duration + deltaTime
+      const next = applyResizeEndRules(track.clips, clipId, desiredEnd, currentTime)
+      // sourceDuration clamp (if known)
+      const sd = clip.sourceDuration
+      const trimStart = clip.trimStart ?? 0
+      let duration = next.duration
+      if (typeof sd === 'number' && sd > 0) {
+        duration = Math.min(duration, Math.max(0.5, sd - trimStart))
+      }
+      const trimEnd = trimStart + duration
+
+      // ripple push following clips when extending
+      if (track.type === 'video' && rippleMode) {
+        const ordered = track.clips.slice().sort((a, b) => a.start - b.start)
+        const idx = ordered.findIndex(c => c.id === clipId)
+        const oldEnd = clip.start + clip.duration
+        const newEnd = clip.start + duration
+        if (idx !== -1 && newEnd > oldEnd + 1e-6 && idx + 1 < ordered.length) {
+          const nextClip = ordered[idx + 1]
+          const overlap = newEnd - nextClip.start
+          if (overlap > 1e-6) {
+            rippleShiftFrom(track.id, nextClip.id, overlap)
+          }
+        }
+      }
+
+      updateClip(track.id, clipId, { duration, trimEnd })
+    }
+  }, [track, updateClip, currentTime, rippleMode, rippleShiftFrom])
+  
+  return (
+    <div
+      className={cn(
+        'absolute left-0 right-0 border-b border-editor-border',
+        track.locked && 'opacity-50',
+        track.muted && 'opacity-60'
+      )}
+      style={{ top, height: trackHeight }}
+    >
+      {/* 背景网格 */}
+      <div className="absolute inset-0 bg-editor-bg" />
+      
+      {/* 片段 */}
+      {track.clips
+        .filter((clip) => {
+          if (typeof viewStartTime !== 'number' || typeof viewEndTime !== 'number') return true
+          const start = clip.start ?? 0
+          const end = start + (clip.duration ?? 0)
+          return end >= viewStartTime && start <= viewEndTime
+        })
+        .map(clip => (
+        <ClipItem
+          key={clip.id}
+          clip={clip}
+          trackType={track.type}
+          pixelsPerSecond={pixelsPerSecond}
+          trackHeight={trackHeight}
+          isSelected={selectedClipIds.includes(clip.id)}
+          isLocked={track.locked}
+          onSelect={(e) => {
+            const isToggle = e.shiftKey || e.metaKey || e.ctrlKey
+            toggleClipSelection(clip.id, isToggle ? 'toggle' : 'replace')
+          }}
+          onDrag={(deltaX, deltaTime) => handleClipDrag(clip.id, deltaX, deltaTime)}
+          onResize={(edge, deltaTime) => handleClipResize(clip.id, edge, deltaTime)}
+          onCommit={() => {
+            setSnapGuideTime(null)
+            onClipChange?.()
+          }}
+        />
+      ))}
+    </div>
+  )
+}
+
+export default TrackRow
+

web/src/components/Timeline/index.ts

@@ -0,0 +1,23 @@
+/**
+ * Timeline 组件导出
+ */
+export { Timeline } from './Timeline'
+export { TrackRow } from './TrackRow'
+export { ClipItem } from './ClipItem'
+export { TimeRuler } from './TimeRuler'
+export { Playhead } from './Playhead'
+export { TrackPanel } from './TrackPanel'
+
+
+
+
+
+
+
+
+
+
+
+
+
+

web/src/editorStore.ts

@@ -0,0 +1,5 @@
+// Back-compat re-export
+export * from './store/editorStore'
+
+
+

web/src/index.css

@@ -0,0 +1,118 @@
+@tailwind base;
+@tailwind components;
+@tailwind utilities;
+
+* {
+  margin: 0;
+  padding: 0;
+  box-sizing: border-box;
+}
+
+html, body, #root {
+  height: 100%;
+  width: 100%;
+  overflow: hidden;
+}
+
+body {
+  font-family: 'Inter', system-ui, -apple-system, sans-serif;
+  background-color: #1a1a1a;
+  color: #e5e5e5;
+}
+
+/* 自定义滚动条 */
+::-webkit-scrollbar {
+  width: 8px;
+  height: 8px;
+}
+
+::-webkit-scrollbar-track {
+  background: #242424;
+}
+
+::-webkit-scrollbar-thumb {
+  background: #404040;
+  border-radius: 4px;
+}
+
+::-webkit-scrollbar-thumb:hover {
+  background: #525252;
+}
+
+/* 时间轴拖拽样式 */
+.timeline-clip {
+  cursor: grab;
+  transition: transform 0.1s ease;
+}
+
+.timeline-clip:active {
+  cursor: grabbing;
+}
+
+.timeline-clip.dragging {
+  opacity: 0.8;
+  transform: scale(1.02);
+  z-index: 100;
+}
+
+/* 轨道颜色条 */
+.track-indicator {
+  width: 4px;
+  border-radius: 2px;
+}
+
+/* 播放头 */
+.playhead {
+  position: absolute;
+  top: 0;
+  bottom: 0;
+  width: 2px;
+  background: #ef4444;
+  z-index: 50;
+  pointer-events: none;
+}
+
+.playhead::before {
+  content: '';
+  position: absolute;
+  top: 0;
+  left: 50%;
+  transform: translateX(-50%);
+  width: 0;
+  height: 0;
+  border-left: 6px solid transparent;
+  border-right: 6px solid transparent;
+  border-top: 8px solid #ef4444;
+}
+
+/* 波形容器 */
+.waveform-container {
+  height: 48px;
+  background: rgba(0, 0, 0, 0.2);
+  border-radius: 4px;
+  overflow: hidden;
+}
+
+/* 动画 */
+@keyframes pulse {
+  0%, 100% { opacity: 1; }
+  50% { opacity: 0.5; }
+}
+
+.animate-pulse-slow {
+  animation: pulse 2s ease-in-out infinite;
+}
+
+
+
+
+
+
+
+
+
+
+
+
+
+

web/src/lib/api.ts

@@ -0,0 +1,283 @@
+import axios from 'axios'
+
+// API 客户端配置
+const api = axios.create({
+  baseURL: '/api',
+  timeout: 60000,
+  headers: {
+    'Content-Type': 'application/json',
+  },
+})
+
+// 类型定义
+export interface Project {
+  id: string
+  name: string
+  status: string
+  product_info: Record<string, unknown>
+  script_data?: ScriptData
+  created_at: number
+  updated_at: number
+}
+
+export interface ProjectListItem {
+  id: string
+  name: string
+  status: string
+  updated_at: number
+}
+
+export interface ScriptData {
+  scenes: Scene[]
+  voiceover_timeline: VoiceoverItem[]
+  selling_points?: string[]
+  target_audience?: string
+  bgm_style?: string
+  visual_anchor?: string
+}
+
+export interface Scene {
+  id: number
+  visual_prompt?: string
+  video_prompt?: string
+  fancy_text?: {
+    text: string
+    start_time?: number
+    duration?: number
+  }
+}
+
+export interface VoiceoverItem {
+  text: string
+  subtitle?: string
+  start_time: number
+  duration: number
+}
+
+export interface TimelineClip {
+  id: string
+  type: 'video' | 'audio' | 'subtitle' | 'fancy_text' | 'bgm' | 'sticker'
+  start: number
+  duration: number
+  source_path?: string
+  source_url?: string
+  trim_start?: number
+  trim_end?: number
+  source_duration?: number
+  text?: string
+  style?: Record<string, unknown>
+  position?: { x: string | number; y: string | number }
+  volume?: number
+  fade_in?: number
+  fade_out?: number
+  ducking?: boolean
+  duck_volume?: number
+  playback_rate?: number
+}
+
+export interface Track {
+  id: string
+  name: string
+  type: string
+  clips: TimelineClip[]
+  locked: boolean
+  visible: boolean
+  muted: boolean
+}
+
+export interface EditorState {
+  project_id: string
+  total_duration: number
+  tracks: Track[]
+  current_time: number
+  zoom: number
+  ripple_mode?: boolean
+  subtitle_style?: Record<string, unknown>
+}
+
+export interface BGMItem {
+  id: string
+  name: string
+  path: string
+  url: string
+}
+
+export interface StickerItem {
+  id: string
+  name: string
+  url: string
+  kind?: 'builtin' | 'custom'
+  tags?: string[]
+  category?: string
+  license?: string | null
+  attribution?: string | null
+}
+
+export interface VoiceOption {
+  id: string
+  name: string
+}
+
+// API 方法
+
+// 将前端 store 的 clip/track（camelCase）转换为后端接口期望的 snake_case
+const toApiClip = (clip: any) => ({
+  ...clip,
+  source_path: clip.source_path ?? clip.sourcePath ?? clip.source_path,
+  source_url: clip.source_url ?? clip.sourceUrl ?? clip.source_url,
+  trim_start: clip.trim_start ?? clip.trimStart ?? clip.trim_start ?? 0,
+  trim_end: clip.trim_end ?? clip.trimEnd ?? clip.trim_end ?? null,
+  source_duration: clip.source_duration ?? clip.sourceDuration ?? clip.source_duration ?? null,
+  fade_in: clip.fade_in ?? clip.fadeIn ?? null,
+  fade_out: clip.fade_out ?? clip.fadeOut ?? null,
+  ducking: typeof clip.ducking === 'boolean' ? clip.ducking : null,
+  duck_volume: clip.duck_volume ?? clip.duckVolume ?? null,
+  playback_rate: clip.playback_rate ?? clip.playbackRate ?? null,
+})
+
+const toApiTrack = (track: any) => ({
+  ...track,
+  clips: Array.isArray(track.clips) ? track.clips.map(toApiClip) : [],
+})
+
+// Projects
+export const projectsApi = {
+  list: () => api.get<ProjectListItem[]>('/projects').then(r => r.data),
+  
+  get: (id: string) => api.get<Project>(`/projects/${id}`).then(r => r.data),
+  
+  create: (name: string, productInfo: Record<string, string>) =>
+    api.post('/projects', { name, product_info: productInfo }).then(r => r.data),
+  
+  getAssets: (id: string) => 
+    api.get(`/projects/${id}/assets`).then(r => r.data),
+    
+  generateScript: (id: string, modelProvider = 'shubiaobiao') =>
+    api.post(`/projects/${id}/generate-script`, null, { params: { model_provider: modelProvider } }).then(r => r.data),
+    
+  generateImages: (id: string, modelProvider = 'shubiaobiao') =>
+    api.post(`/projects/${id}/generate-images`, null, { params: { model_provider: modelProvider } }).then(r => r.data),
+    
+  generateVideos: (id: string) =>
+    api.post(`/projects/${id}/generate-videos`).then(r => r.data),
+}
+
+// Editor
+export const editorApi = {
+  getState: (projectId: string) =>
+    api.get<EditorState>(`/editor/${projectId}/state`).then(r => r.data),
+    
+  saveState: (projectId: string, state: EditorState) =>
+    api.post(`/editor/${projectId}/state`, {
+      ...state,
+      tracks: (state as any).tracks?.map(toApiTrack) ?? [],
+    }).then(r => r.data),
+    
+  generateVoiceover: (text: string, voiceType?: string, targetDuration?: number) =>
+    api.post('/editor/generate-voiceover', {
+      text,
+      voice_type: voiceType,
+      target_duration: targetDuration,
+    }).then(r => r.data),
+    
+  generateFancyText: (text: string, style?: Record<string, unknown>) =>
+    api.post('/editor/generate-fancy-text', { text, style }).then(r => r.data),
+    
+  trimVideo: (sourcePath: string, startTime: number, endTime: number) =>
+    api.post('/editor/trim-video', {
+      source_path: sourcePath,
+      start_time: startTime,
+      end_time: endTime,
+    }).then(r => r.data),
+    
+  deleteClip: (projectId: string, clipId: string) =>
+    api.delete(`/editor/${projectId}/clip/${clipId}`).then(r => r.data),
+}
+
+// Compose
+export const composeApi = {
+  render: (projectId: string, tracks: Track[], options?: {
+    voiceType?: string
+    bgmVolume?: number
+    outputName?: string
+  }) => {
+    const videoClips = (tracks.find(t => t.type === 'video')?.clips || []).map(toApiClip)
+    const voiceoverClips = (tracks.find(t => t.id === 'audio-voiceover')?.clips || []).map(toApiClip)
+    const subtitleClips = (tracks.find(t => t.type === 'subtitle')?.clips || []).map(toApiClip)
+    const fancyTextClips = (tracks.find(t => t.type === 'fancy_text')?.clips || []).map(toApiClip)
+    const stickerClips = (tracks.find(t => t.type === 'sticker')?.clips || []).map(toApiClip)
+    const bgmClipRaw = tracks.find(t => t.type === 'bgm')?.clips[0] || null
+    const bgmClip = bgmClipRaw ? toApiClip(bgmClipRaw) : null
+    
+    return api.post('/compose/render', {
+      project_id: projectId,
+      video_clips: videoClips,
+      voiceover_clips: voiceoverClips,
+      subtitle_clips: subtitleClips,
+      fancy_text_clips: fancyTextClips,
+      sticker_clips: stickerClips,
+      bgm_clip: bgmClip,
+      voice_type: options?.voiceType || 'zh_female_santongyongns_saturn_bigtts',
+      bgm_volume: options?.bgmVolume || 0.15,
+      output_name: options?.outputName,
+    }).then(r => r.data)
+  },
+  
+  quickCompose: (projectId: string, bgmId?: string) =>
+    api.post('/compose/quick', null, { params: { project_id: projectId, bgm_id: bgmId } }).then(r => r.data),
+    
+  getStatus: (taskId: string) =>
+    api.get(`/compose/status/${taskId}`).then(r => r.data),
+
+  retry: (taskId: string) =>
+    api.post(`/compose/retry/${taskId}`).then(r => r.data),
+}
+
+// Assets
+export const assetsApi = {
+  list: (projectId: string, assetType?: string) =>
+    api.get('/assets/list/' + projectId, { params: { asset_type: assetType } }).then(r => r.data),
+    
+  upload: (file: File, assetType = 'custom') => {
+    const formData = new FormData()
+    formData.append('file', file)
+    formData.append('asset_type', assetType)
+    return api.post('/assets/upload', formData, {
+      headers: { 'Content-Type': 'multipart/form-data' },
+    }).then(r => r.data)
+  },
+  
+  getBGM: () => api.get<BGMItem[]>('/assets/bgm').then(r => r.data),
+  
+  getFonts: () => api.get('/assets/fonts').then(r => r.data),
+
+  uploadFont: (file: File) => {
+    const formData = new FormData()
+    formData.append('file', file)
+    return api.post('/assets/fonts/upload', formData, {
+      headers: { 'Content-Type': 'multipart/form-data' },
+    }).then(r => r.data)
+  },
+
+  getStickers: () => api.get<StickerItem[]>('/assets/stickers').then(r => r.data),
+
+  uploadSticker: (file: File) => {
+    const formData = new FormData()
+    formData.append('file', file)
+    return api.post('/assets/stickers/upload', formData, {
+      headers: { 'Content-Type': 'multipart/form-data' },
+    }).then(r => r.data)
+  },
+}
+
+// Config
+export const configApi = {
+  get: () => api.get('/config').then(r => r.data),
+  health: () => api.get('/health').then(r => r.data),
+}
+
+export default api
+
+
+
+

web/src/lib/templates.ts

@@ -0,0 +1,86 @@
+/**
+ * 轻量模板系统（MVP）
+ * - 目标：把“模板=一组默认样式+可选片段生成规则”固化，后续可演进为 zip 包
+ */
+import type { Track, TimelineClip } from '@/store/editorStore'
+import { generateId } from '@/lib/utils'
+
+export type TemplateApplyResult = {
+  subtitleStyle?: Record<string, unknown>
+  addClips?: Array<{ trackType: string; clip: Omit<TimelineClip, 'id'> }>
+}
+
+export interface EditorTemplate {
+  id: string
+  name: string
+  description: string
+  apply: (ctx: { tracks: Track[]; totalDuration: number }) => TemplateApplyResult
+}
+
+const ensureTrack = (tracks: Track[], type: string) => tracks.find(t => t.type === type)
+
+export const templates: EditorTemplate[] = [
+  {
+    id: 'tpl_subtitle_box',
+    name: '字幕-底部半透明底',
+    description: '适合口播：底部字幕带半透明底，描边较轻。',
+    apply: () => ({
+      subtitleStyle: {
+        fontsize: 56,
+        fontcolor: 'white',
+        borderw: 3,
+        bordercolor: 'black',
+        box: 1,
+        boxcolor: 'black@0.45',
+        boxborderw: 18,
+        x: '(w-text_w)/2',
+        y: 'h-220',
+      },
+    }),
+  },
+  {
+    id: 'tpl_intro_outro_fancy',
+    name: '花字-片头片尾',
+    description: '自动加一个片头/片尾花字（可后续手动改文本/位置）。',
+    apply: ({ tracks, totalDuration }) => {
+      const fancy = ensureTrack(tracks, 'fancy_text')
+      if (!fancy) return {}
+      const d = Math.max(0.8, Math.min(1.6, totalDuration / 8))
+      const intro: Omit<TimelineClip, 'id'> = {
+        type: 'fancy_text',
+        start: 0,
+        duration: d,
+        text: '片头标题',
+        style: { font_size: 80, font_color: '#FFFFFF' },
+        position: { x: '50%', y: '160px' },
+      }
+      const outro: Omit<TimelineClip, 'id'> = {
+        type: 'fancy_text',
+        start: Math.max(0, totalDuration - d),
+        duration: d,
+        text: '片尾收束',
+        style: { font_size: 72, font_color: '#FFFFFF' },
+        position: { x: '50%', y: '160px' },
+      }
+      return {
+        addClips: [
+          { trackType: 'fancy_text', clip: intro },
+          { trackType: 'fancy_text', clip: outro },
+        ],
+      }
+    },
+  },
+]
+
+export const applyTemplateToTracks = (tracks: Track[], res: TemplateApplyResult) => {
+  const ops: Array<{ trackId: string; clip: TimelineClip }> = []
+  for (const add of res.addClips || []) {
+    const t = tracks.find(x => x.type === add.trackType)
+    if (!t) continue
+    ops.push({ trackId: t.id, clip: { ...add.clip, id: generateId() } as TimelineClip })
+  }
+  return ops
+}
+
+
+

web/src/lib/timelineRules.ts

@@ -0,0 +1,210 @@
+/**
+ * Timeline rules layer (MVP)
+ * - snapping: playhead, other clip edges, grid
+ * - collision: disallow overlap in same track
+ * - normalize: keep trimEnd = trimStart + duration
+ */
+import type { TimelineClip } from '@/store/editorStore'
+
+export interface TimelineRuleOptions {
+  gridSeconds?: number
+  snapThresholdSeconds?: number
+  minDurationSeconds?: number
+}
+
+const DEFAULTS: Required<TimelineRuleOptions> = {
+  gridSeconds: 0.5,
+  snapThresholdSeconds: 0.12,
+  minDurationSeconds: 0.5,
+}
+
+const clamp = (v: number, min: number, max: number) => Math.max(min, Math.min(max, v))
+
+export const normalizeTrim = (clip: TimelineClip): TimelineClip => {
+  const trimStart = clip.trimStart ?? 0
+  const duration = clip.duration
+  const trimEnd = trimStart + duration
+  return { ...clip, trimStart, trimEnd }
+}
+
+export const buildSnapAnchors = (
+  clips: TimelineClip[],
+  excludeClipId: string,
+  playheadTime: number,
+  opts?: TimelineRuleOptions
+) => {
+  const o = { ...DEFAULTS, ...(opts || {}) }
+  const anchors = new Set<number>()
+  anchors.add(playheadTime)
+
+  // grid near playhead, and general
+  const grid = o.gridSeconds
+  anchors.add(Math.round(playheadTime / grid) * grid)
+
+  // other clip edges
+  for (const c of clips) {
+    if (c.id === excludeClipId) continue
+    anchors.add(c.start)
+    anchors.add(c.start + c.duration)
+  }
+  return { anchors: Array.from(anchors), opts: o }
+}
+
+export const snapTime = (time: number, anchors: number[], opts?: TimelineRuleOptions) => {
+  const o = { ...DEFAULTS, ...(opts || {}) }
+  let best = time
+  let bestDist = Infinity
+  for (const a of anchors) {
+    const d = Math.abs(a - time)
+    if (d < bestDist) {
+      bestDist = d
+      best = a
+    }
+  }
+  return bestDist <= o.snapThresholdSeconds ? best : time
+}
+
+export const resolveNoOverlapStart = (
+  clips: TimelineClip[],
+  clipId: string,
+  desiredStart: number,
+  duration: number
+) => {
+  const others = clips.filter(c => c.id !== clipId).slice().sort((a, b) => a.start - b.start)
+  const desiredEnd = desiredStart + duration
+
+  // find nearest prev/next boundaries
+  let prevEnd = 0
+  let nextStart = Number.POSITIVE_INFINITY
+  for (const c of others) {
+    const cStart = c.start
+    const cEnd = c.start + c.duration
+    if (cEnd <= desiredStart) {
+      prevEnd = Math.max(prevEnd, cEnd)
+      continue
+    }
+    if (cStart >= desiredEnd) {
+      nextStart = Math.min(nextStart, cStart)
+      break
+    }
+    // overlap with c
+    if (cStart < desiredEnd && cEnd > desiredStart) {
+      // clamp to nearest edge (prefer not moving backward too much)
+      // If desiredStart is before cStart, clamp end to cStart; else clamp start to cEnd.
+      const clampToBefore = cStart - duration
+      const clampToAfter = cEnd
+      const distBefore = Math.abs(desiredStart - clampToBefore)
+      const distAfter = Math.abs(desiredStart - clampToAfter)
+      return distBefore <= distAfter ? Math.max(0, clampToBefore) : clampToAfter
+    }
+  }
+  const maxStart = nextStart === Number.POSITIVE_INFINITY ? Number.POSITIVE_INFINITY : nextStart - duration
+  return clamp(desiredStart, prevEnd, maxStart)
+}
+
+export const applyDragRules = (
+  clips: TimelineClip[],
+  clipId: string,
+  desiredStart: number,
+  playheadTime: number,
+  opts?: TimelineRuleOptions
+) => {
+  const clip = clips.find(c => c.id === clipId)
+  if (!clip) return desiredStart
+  const { anchors, opts: o } = buildSnapAnchors(clips, clipId, playheadTime, opts)
+  let nextStart = Math.max(0, desiredStart)
+  nextStart = snapTime(nextStart, anchors, o)
+  nextStart = resolveNoOverlapStart(clips, clipId, nextStart, clip.duration)
+  return nextStart
+}
+
+export const applyResizeStartRules = (
+  clips: TimelineClip[],
+  clipId: string,
+  desiredStart: number,
+  playheadTime: number,
+  opts?: TimelineRuleOptions
+) => {
+  const o = { ...DEFAULTS, ...(opts || {}) }
+  const clip = clips.find(c => c.id === clipId)
+  if (!clip) {
+    const duration = o.minDurationSeconds
+    const trimStart = 0
+    const trimEnd = trimStart + duration
+    return { start: Math.max(0, desiredStart), duration, trimStart, trimEnd }
+  }
+
+  const oldEnd = clip.start + clip.duration
+  const minStart = Math.max(0, oldEnd - o.minDurationSeconds)
+
+  const { anchors } = buildSnapAnchors(clips, clipId, playheadTime, o)
+  let start = clamp(desiredStart, 0, minStart)
+  start = snapTime(start, anchors, o)
+
+  // collision: ensure new [start, oldEnd] doesn't overlap others
+  // For start-resize we clamp start to be >= prevEnd and <= nextStart - minDuration
+  const others = clips.filter(c => c.id !== clipId).slice().sort((a, b) => a.start - b.start)
+  let prevEnd = 0
+  let nextStart = Number.POSITIVE_INFINITY
+  for (const c of others) {
+    const cEnd = c.start + c.duration
+    if (cEnd <= start) {
+      prevEnd = Math.max(prevEnd, cEnd)
+      continue
+    }
+    if (c.start >= oldEnd) {
+      nextStart = Math.min(nextStart, c.start)
+      break
+    }
+    // overlapping other clip within [start, oldEnd]
+    if (c.start < oldEnd && cEnd > start) {
+      start = cEnd
+    }
+  }
+  const maxStart = Math.min(minStart, nextStart - o.minDurationSeconds)
+  start = clamp(start, prevEnd, maxStart)
+
+  // IMPORTANT: duration must be recalculated after collision adjustments,
+  // otherwise right edge would drift and can overlap next clip.
+  const duration = Math.max(o.minDurationSeconds, oldEnd - start)
+
+  const startDiff = start - clip.start
+  const trimStart = (clip.trimStart ?? 0) + startDiff
+  const trimEnd = trimStart + duration
+
+  return { start, duration, trimStart, trimEnd }
+}
+
+export const applyResizeEndRules = (
+  clips: TimelineClip[],
+  clipId: string,
+  desiredEnd: number,
+  playheadTime: number,
+  opts?: TimelineRuleOptions
+) => {
+  const o = { ...DEFAULTS, ...(opts || {}) }
+  const clip = clips.find(c => c.id === clipId)
+  if (!clip) return { duration: o.minDurationSeconds, trimEnd: o.minDurationSeconds }
+
+  const { anchors } = buildSnapAnchors(clips, clipId, playheadTime, o)
+  let end = Math.max(clip.start + o.minDurationSeconds, desiredEnd)
+  end = snapTime(end, anchors, o)
+
+  // collision: ensure end <= nextStart
+  const others = clips.filter(c => c.id !== clipId).slice().sort((a, b) => a.start - b.start)
+  for (const c of others) {
+    if (c.start >= clip.start + o.minDurationSeconds) {
+      if (c.start >= clip.start && c.start < end) {
+        end = Math.min(end, c.start)
+      }
+      if (c.start >= end) break
+    }
+  }
+
+  const duration = Math.max(o.minDurationSeconds, end - clip.start)
+  const trimStart = clip.trimStart ?? 0
+  const trimEnd = trimStart + duration
+  return { duration, trimEnd }
+}
+
+

web/src/lib/transitions.ts

@@ -0,0 +1,69 @@
+export type TransitionCategory =
+  | '基础'
+  | '缩放'
+  | '滑动'
+  | '旋转'
+  | '模糊'
+  | '光效'
+  | '颜色'
+
+export type TransitionPreset = {
+  id: string
+  name: string
+  desc: string
+  category: TransitionCategory
+  tags?: string[]
+  // 映射到 clip.style
+  type: string
+  defaultDurationSec: number
+  // 导出是否支持（MVP：先只保证 fade 一致）
+  exportable: boolean
+}
+
+/**
+ * 转场库（火山式：不改片段时长、不插入片段）
+ * - 预览：Remotion 末尾动效
+ * - 导出：逐步补齐（先保证 fade）
+ */
+export const transitionCatalog: TransitionPreset[] = [
+  // 基础
+  { id: 'fade', name: '淡出', desc: '片尾淡出到黑（导出支持）', category: '基础', type: 'fade', defaultDurationSec: 0.6, exportable: true, tags: ['通用', '稳'] },
+  { id: 'dipToBlack', name: '快速淡黑', desc: '更短更利落的淡出', category: '基础', type: 'fade', defaultDurationSec: 0.25, exportable: true, tags: ['节奏', '快'] },
+  { id: 'fadeWhite', name: '淡到白', desc: '片尾淡出到白（导出支持）', category: '基础', type: 'fadeWhite', defaultDurationSec: 0.5, exportable: true, tags: ['明亮'] },
+  { id: 'flash', name: '闪白', desc: '片尾闪一下（导出支持）', category: '光效', type: 'flash', defaultDurationSec: 0.25, exportable: true, tags: ['节奏'] },
+
+  // 缩放
+  { id: 'zoomOut', name: '缩小', desc: '片尾逐渐缩小（导出支持）', category: '缩放', type: 'zoomOut', defaultDurationSec: 0.6, exportable: true, tags: ['动感'] },
+  { id: 'zoomIn', name: '推进', desc: '片尾逐渐推进（导出支持）', category: '缩放', type: 'zoomIn', defaultDurationSec: 0.6, exportable: true, tags: ['动感'] },
+
+  // 滑动
+  { id: 'slideLeft', name: '左滑', desc: '片尾向左滑走（导出支持）', category: '滑动', type: 'slideLeft', defaultDurationSec: 0.6, exportable: true, tags: ['通用'] },
+  { id: 'slideRight', name: '右滑', desc: '片尾向右滑走（导出支持）', category: '滑动', type: 'slideRight', defaultDurationSec: 0.6, exportable: true, tags: ['通用'] },
+  { id: 'slideUp', name: '上滑', desc: '片尾向上滑走（导出支持）', category: '滑动', type: 'slideUp', defaultDurationSec: 0.6, exportable: true, tags: ['通用'] },
+  { id: 'slideDown', name: '下滑', desc: '片尾向下滑走（导出支持）', category: '滑动', type: 'slideDown', defaultDurationSec: 0.6, exportable: true, tags: ['通用'] },
+
+  // 旋转
+  { id: 'rotateOut', name: '旋转', desc: '片尾小角度旋转（导出支持）', category: '旋转', type: 'rotateOut', defaultDurationSec: 0.6, exportable: true, tags: ['动感'] },
+
+  // 模糊
+  { id: 'blurOut', name: '模糊', desc: '片尾逐渐模糊（导出支持）', category: '模糊', type: 'blurOut', defaultDurationSec: 0.6, exportable: true, tags: ['质感'] },
+  { id: 'blurFade', name: '模糊淡出', desc: '片尾模糊并淡出（导出支持）', category: '模糊', type: 'blurFade', defaultDurationSec: 0.6, exportable: true, tags: ['质感', '通用'] },
+
+  // 颜色
+  { id: 'desaturate', name: '去饱和', desc: '片尾逐渐变灰（导出支持）', category: '颜色', type: 'desaturate', defaultDurationSec: 0.6, exportable: true, tags: ['质感'] },
+  { id: 'colorPop', name: '色彩增强', desc: '片尾增强饱和/对比（导出支持）', category: '颜色', type: 'colorPop', defaultDurationSec: 0.6, exportable: true, tags: ['动感'] },
+  { id: 'hueShift', name: '色相偏移', desc: '片尾轻微色相旋转（导出支持）', category: '颜色', type: 'hueShift', defaultDurationSec: 0.6, exportable: true, tags: ['氛围'] },
+  { id: 'darken', name: '变暗', desc: '片尾逐渐变暗（导出支持）', category: '颜色', type: 'darken', defaultDurationSec: 0.6, exportable: true, tags: ['氛围'] },
+]
+
+export const transitionCategories: TransitionCategory[] = [
+  '基础',
+  '缩放',
+  '滑动',
+  '旋转',
+  '模糊',
+  '光效',
+  '颜色',
+]
+
+

web/src/lib/utils.ts

@@ -0,0 +1,160 @@
+import { clsx, type ClassValue } from 'clsx'
+import { twMerge } from 'tailwind-merge'
+
+export function cn(...inputs: ClassValue[]) {
+  return twMerge(clsx(inputs))
+}
+
+/**
+ * 格式化时间（秒）为 MM:SS.ms 格式
+ */
+export function formatTime(seconds: number): string {
+  const mins = Math.floor(seconds / 60)
+  const secs = Math.floor(seconds % 60)
+  const ms = Math.floor((seconds % 1) * 100)
+  return `${mins.toString().padStart(2, '0')}:${secs.toString().padStart(2, '0')}.${ms.toString().padStart(2, '0')}`
+}
+
+/**
+ * 格式化时间为简短格式 M:SS
+ */
+export function formatTimeShort(seconds: number): string {
+  const mins = Math.floor(seconds / 60)
+  const secs = Math.floor(seconds % 60)
+  return `${mins}:${secs.toString().padStart(2, '0')}`
+}
+
+/**
+ * 解析用户输入时间到秒。
+ * 支持：
+ * - "M:SS" / "MM:SS"
+ * - "M:SS.ms" / "MM:SS.ms"
+ * - 纯数字秒（如 "12.5"）
+ */
+export function parseTimeInput(input: string): number | null {
+  const raw = (input || '').trim()
+  if (!raw) return null
+
+  // pure number seconds
+  if (/^\d+(\.\d+)?$/.test(raw)) {
+    const n = Number(raw)
+    return Number.isFinite(n) ? n : null
+  }
+
+  // M:SS(.ms)
+  const m = raw.match(/^(\d+)\s*:\s*(\d{1,2})(?:\.(\d{1,3}))?$/)
+  if (!m) return null
+  const mins = Number(m[1])
+  const secs = Number(m[2])
+  const fracRaw = m[3]
+  if (!Number.isFinite(mins) || !Number.isFinite(secs) || secs >= 60) return null
+  let frac = 0
+  if (fracRaw) {
+    const fracN = Number(`0.${fracRaw.padEnd(3, '0')}`) // treat as ms-ish
+    frac = Number.isFinite(fracN) ? fracN : 0
+  }
+  return mins * 60 + secs + frac
+}
+
+/**
+ * 将像素转换为时间（基于缩放比例）
+ */
+export function pixelsToTime(pixels: number, pixelsPerSecond: number): number {
+  return pixels / pixelsPerSecond
+}
+
+/**
+ * 将时间转换为像素（基于缩放比例）
+ */
+export function timeToPixels(time: number, pixelsPerSecond: number): number {
+  return time * pixelsPerSecond
+}
+
+/**
+ * 限制数值在范围内
+ */
+export function clamp(value: number, min: number, max: number): number {
+  return Math.min(Math.max(value, min), max)
+}
+
+/**
+ * 生成唯一 ID
+ */
+export function generateId(): string {
+  return `${Date.now()}-${Math.random().toString(36).substr(2, 9)}`
+}
+
+/**
+ * 节流函数
+ */
+export function throttle<T extends (...args: unknown[]) => void>(
+  func: T,
+  limit: number
+): T {
+  let inThrottle: boolean
+  return function(this: unknown, ...args: Parameters<T>) {
+    if (!inThrottle) {
+      func.apply(this, args)
+      inThrottle = true
+      setTimeout(() => (inThrottle = false), limit)
+    }
+  } as T
+}
+
+/**
+ * 防抖函数
+ */
+export function debounce<T extends (...args: unknown[]) => void>(
+  func: T,
+  wait: number
+): T {
+  let timeout: ReturnType<typeof setTimeout>
+  return function(this: unknown, ...args: Parameters<T>) {
+    clearTimeout(timeout)
+    timeout = setTimeout(() => func.apply(this, args), wait)
+  } as T
+}
+
+/**
+ * 获取轨道类型的颜色
+ */
+export function getTrackColor(type: string): string {
+  const colors: Record<string, string> = {
+    video: 'bg-track-video',
+    audio: 'bg-track-audio',
+    voiceover: 'bg-track-voiceover',
+    subtitle: 'bg-track-subtitle',
+    fancy_text: 'bg-track-fancy',
+    bgm: 'bg-track-bgm',
+  }
+  return colors[type] || 'bg-gray-500'
+}
+
+/**
+ * 获取轨道类型的边框颜色
+ */
+export function getTrackBorderColor(type: string): string {
+  const colors: Record<string, string> = {
+    video: 'border-track-video',
+    audio: 'border-track-audio',
+    voiceover: 'border-track-voiceover',
+    subtitle: 'border-track-subtitle',
+    fancy_text: 'border-track-fancy',
+    bgm: 'border-track-bgm',
+  }
+  return colors[type] || 'border-gray-500'
+}
+
+
+
+
+
+
+
+
+
+
+
+
+
+

web/src/main.tsx

@@ -0,0 +1,36 @@
+import React from 'react'
+import ReactDOM from 'react-dom/client'
+import { QueryClient, QueryClientProvider } from '@tanstack/react-query'
+import App from './App'
+import './index.css'
+
+const queryClient = new QueryClient({
+  defaultOptions: {
+    queries: {
+      staleTime: 1000 * 60, // 1 minute
+      refetchOnWindowFocus: false,
+    },
+  },
+})
+
+ReactDOM.createRoot(document.getElementById('root')!).render(
+  <React.StrictMode>
+    <QueryClientProvider client={queryClient}>
+      <App />
+    </QueryClientProvider>
+  </React.StrictMode>,
+)
+
+
+
+
+
+
+
+
+
+
+
+
+
+

web/src/pages/ProjectsPage.tsx

@@ -0,0 +1,172 @@
+/**
+ * 项目列表页面
+ */
+import React, { useMemo } from 'react'
+import { useNavigate } from 'react-router-dom'
+import { useQuery } from '@tanstack/react-query'
+import { Plus, Film, Clock, ChevronRight } from 'lucide-react'
+import { projectsApi } from '@/lib/api'
+import { cn } from '@/lib/utils'
+
+export const ProjectsPage: React.FC = () => {
+  const navigate = useNavigate()
+  const workflowUrl = useMemo(() => {
+    // 生产环境部署在同一台机器：前端 3000，控制台 8503
+    try {
+      const u = new URL(window.location.href)
+      return `${u.protocol}//${u.hostname}:8503`
+    } catch {
+      return 'http://localhost:8503'
+    }
+  }, [])
+  
+  const { data: projects, isLoading, error } = useQuery({
+    queryKey: ['projects'],
+    queryFn: projectsApi.list,
+  })
+  
+  const formatDate = (timestamp: number) => {
+    return new Date(timestamp * 1000).toLocaleString('zh-CN', {
+      month: 'short',
+      day: 'numeric',
+      hour: '2-digit',
+      minute: '2-digit',
+    })
+  }
+  
+  const getStatusBadge = (status: string) => {
+    const styles: Record<string, string> = {
+      created: 'bg-gray-500/20 text-gray-400',
+      script_generated: 'bg-blue-500/20 text-blue-400',
+      images_generated: 'bg-purple-500/20 text-purple-400',
+      videos_generated: 'bg-cyan-500/20 text-cyan-400',
+      completed: 'bg-green-500/20 text-green-400',
+      failed: 'bg-red-500/20 text-red-400',
+    }
+    
+    const labels: Record<string, string> = {
+      created: '已创建',
+      script_generated: '脚本完成',
+      images_generated: '图片完成',
+      videos_generated: '视频完成',
+      completed: '已完成',
+      failed: '失败',
+    }
+    
+    return (
+      <span className={cn(
+        'px-2 py-0.5 rounded text-xs font-medium',
+        styles[status] || styles.created
+      )}>
+        {labels[status] || status}
+      </span>
+    )
+  }
+  
+  return (
+    <div className="min-h-screen bg-editor-bg">
+      {/* Header */}
+      <header className="bg-editor-panel border-b border-editor-border">
+        <div className="max-w-6xl mx-auto px-6 py-4 flex items-center justify-between">
+          <div className="flex items-center gap-3">
+            <Film className="w-8 h-8 text-editor-accent" />
+            <h1 className="text-xl font-bold text-editor-text">Video Flow Editor</h1>
+          </div>
+          
+          <a
+            href={workflowUrl}
+            target="_blank"
+            rel="noopener noreferrer"
+            className="text-sm text-editor-text-muted hover:text-editor-text transition-colors"
+          >
+            打开工作流控制台 →
+          </a>
+        </div>
+      </header>
+      
+      {/* Content */}
+      <main className="max-w-6xl mx-auto px-6 py-8">
+        <div className="flex items-center justify-between mb-6">
+          <h2 className="text-lg font-semibold text-editor-text">我的项目</h2>
+          
+          <a
+            href={workflowUrl}
+            target="_blank"
+            rel="noopener noreferrer"
+            className={cn(
+              'flex items-center gap-2 px-4 py-2 rounded-lg',
+              'bg-editor-accent text-white font-medium',
+              'hover:bg-editor-accent-hover transition-colors'
+            )}
+          >
+            <Plus className="w-4 h-4" />
+            新建项目
+          </a>
+        </div>
+        
+        {isLoading ? (
+          <div className="text-center py-12">
+            <div className="animate-spin w-8 h-8 border-2 border-editor-accent border-t-transparent rounded-full mx-auto" />
+            <p className="mt-4 text-editor-text-muted">加载中...</p>
+          </div>
+        ) : error ? (
+          <div className="text-center py-12">
+            <p className="text-red-400">加载失败，请检查后端服务是否启动</p>
+            <p className="text-sm text-editor-text-muted mt-2">
+              确保 FastAPI 服务运行在 http://localhost:8000
+            </p>
+          </div>
+        ) : projects?.length === 0 ? (
+          <div className="text-center py-12 bg-editor-panel rounded-xl border border-editor-border">
+            <Film className="w-12 h-12 text-editor-text-muted mx-auto" />
+            <p className="mt-4 text-editor-text-muted">还没有项目</p>
+            <p className="text-sm text-editor-text-muted mt-1">
+              前往工作流控制台创建新项目
+            </p>
+          </div>
+        ) : (
+          <div className="grid gap-4">
+            {projects?.map(project => (
+              <div
+                key={project.id}
+                onClick={() => navigate(`/editor/${project.id}`)}
+                className={cn(
+                  'flex items-center justify-between p-4 rounded-xl',
+                  'bg-editor-panel border border-editor-border',
+                  'hover:border-editor-accent/50 cursor-pointer transition-all',
+                  'group'
+                )}
+              >
+                <div className="flex items-center gap-4">
+                  <div className="w-12 h-12 rounded-lg bg-editor-surface flex items-center justify-center">
+                    <Film className="w-6 h-6 text-editor-text-muted" />
+                  </div>
+                  
+                  <div>
+                    <h3 className="font-medium text-editor-text group-hover:text-editor-accent transition-colors">
+                      {project.name}
+                    </h3>
+                    <div className="flex items-center gap-3 mt-1">
+                      {getStatusBadge(project.status)}
+                      <span className="flex items-center gap-1 text-xs text-editor-text-muted">
+                        <Clock className="w-3 h-3" />
+                        {formatDate(project.updated_at)}
+                      </span>
+                    </div>
+                  </div>
+                </div>
+                
+                <ChevronRight className="w-5 h-5 text-editor-text-muted group-hover:text-editor-accent transition-colors" />
+              </div>
+            ))}
+          </div>
+        )}
+      </main>
+    </div>
+  )
+}
+
+export default ProjectsPage
+
+
+

web/src/pages/index.ts

@@ -0,0 +1,19 @@
+/**
+ * Pages 导出
+ */
+export { ProjectsPage } from './ProjectsPage'
+export { EditorPage } from './EditorPage'
+
+
+
+
+
+
+
+
+
+
+
+
+
+

web/src/remotion/VideoComposition.tsx

@@ -0,0 +1,866 @@
+/**
+ * Remotion 视频合成组件
+ * 用于浏览器端实时预览
+ */
+import React from 'react'
+import {
+  AbsoluteFill,
+  Sequence,
+  Video,
+  Audio,
+  Img,
+  useCurrentFrame,
+  useVideoConfig,
+  interpolate,
+} from 'remotion'
+import type { Track, TimelineClip } from '@/store/editorStore'
+
+const clamp01 = (v: number) => Math.max(0, Math.min(1, v))
+
+const makeFade = (localFrame: number, durationFrames: number, fps: number, fadeIn?: number, fadeOut?: number) => {
+  const fi = Math.max(0, Number(fadeIn || 0))
+  const fo = Math.max(0, Number(fadeOut || 0))
+  const fiF = Math.round(fi * fps)
+  const foF = Math.round(fo * fps)
+  let f = 1
+  if (fiF > 0) f *= clamp01(localFrame / Math.max(1, fiF))
+  if (foF > 0) f *= clamp01((durationFrames - localFrame) / Math.max(1, foF))
+  return clamp01(f)
+}
+
+const resolvePos = (v: any, total: number, fallback: number) => {
+  if (typeof v === 'number' && Number.isFinite(v)) {
+    // 0~1 视为百分比；否则视为像素
+    if (v >= 0 && v <= 1) return v * total
+    return v
+  }
+  return fallback
+}
+
+const normPosPct = (v: any, total: number, fallbackPct: number) => {
+  const n = Number(v)
+  if (typeof n === 'number' && Number.isFinite(n)) {
+    if (n >= 0 && n <= 1) return n
+    if (total > 0) return n / total
+  }
+  return fallbackPct
+}
+
+type UiBridge = {
+  selectedClipId?: string | null
+  onSelectClip?: (clipId: string) => void
+  onUpdateClip?: (clipId: string, updates: Partial<TimelineClip>) => void
+  onPushHistory?: (meta?: { label?: string; icon?: string }) => void
+}
+
+const RenderSizeCtx = React.createContext<{ w: number; h: number }>({ w: 0, h: 0 })
+
+const clamp = (v: number, lo: number, hi: number) => Math.max(lo, Math.min(hi, v))
+
+const getBoxWPct = (style: any, width: number, fallback: number) => {
+  const bwRaw = style?.box_w ?? style?.boxW ?? style?.max_width ?? style?.maxWidth ?? 0
+  const bw = Number(bwRaw) || 0
+  if (bw > 0 && bw <= 1) return bw
+  if (bw > 1 && width > 0) return clamp(bw / width, 0.2, 1.0)
+  return clamp(fallback, 0.2, 1.0)
+}
+
+const toCssFontFamily = (v: any) => {
+  if (typeof v !== 'string' || !v) return undefined
+  // 若是绝对路径（用于导出 renderer/ffmpeg），预览侧用系统字体名回退
+  if (v.includes('/')) return 'PingFang SC'
+  return v
+}
+
+const tailProgress01 = (frame: number, durationFrames: number, fps: number, tailSec: number) => {
+  const dF = Math.max(1, Math.round(Math.max(0.0, tailSec) * fps))
+  const start = Math.max(0, durationFrames - dF)
+  if (frame < start) return 0
+  return clamp01((frame - start) / Math.max(1, dF))
+}
+
+// 视频片段组件（含原声控制）
+const VideoClip: React.FC<{
+  clip: TimelineClip
+  volume?: number
+}> = ({ clip, volume }) => {
+  if (!clip.sourceUrl) return null
+  const { fps } = useVideoConfig()
+  const frame = useCurrentFrame()
+  const durationFrames = Math.max(1, Math.round((clip.duration || 0) * fps))
+  const s: any = clip.style || {}
+  const vFadeIn = Number(s.vFadeIn ?? s.v_fade_in ?? 0) || 0
+  const vFadeOut = Number(s.vFadeOut ?? s.v_fade_out ?? 0) || 0
+  const opacity = makeFade(frame, durationFrames, fps, vFadeIn, vFadeOut)
+
+  // “火山式”转场：不改片段时长，仅对片段末尾做动效（默认 out）
+  const tType = String(s.vTransitionType ?? s.v_transition_type ?? '')
+  const tDur = Number(s.vTransitionDur ?? s.v_transition_dur ?? 0) || 0
+  const p = tailProgress01(frame, durationFrames, fps, tDur)
+  const outOpacity = (() => {
+    if (tType === 'fade' || tType === 'fadeWhite' || tType === 'blurFade') return 1 - p
+    return 1
+  })()
+  const bgColor = tType === 'fadeWhite' ? '#fff' : 'transparent'
+
+  // css filters（按顺序叠加，必须与后端 FFmpeg 的滤镜意图一致）
+  const filterParts: string[] = []
+  if (tType === 'blurOut') filterParts.push(`blur(${p * 10}px)`)
+  if (tType === 'blurFade') filterParts.push(`blur(${p * 8}px)`)
+  if (tType === 'desaturate') filterParts.push(`saturate(${Math.max(0, 1 - 0.9 * p)})`)
+  if (tType === 'colorPop') {
+    filterParts.push(`saturate(${1 + 0.8 * p})`)
+    filterParts.push(`contrast(${1 + 0.3 * p})`)
+  }
+  if (tType === 'darken') filterParts.push(`brightness(${Math.max(0.1, 1 - 0.4 * p)})`)
+  if (tType === 'hueShift') filterParts.push(`hue-rotate(${60 * p}deg)`)
+
+  // flash：brightness 峰值在 p=0.5
+  if (tType === 'flash') {
+    const k = 1 - Math.abs(0.5 - p) * 2
+    const b = 1 + 0.7 * clamp01(k)
+    filterParts.push(`brightness(${b})`)
+  }
+
+  const outFilter = filterParts.length ? filterParts.join(' ') : 'none'
+  const outTransform = (() => {
+    const parts: string[] = []
+    // slide：像素位移（与导出保持同量级）
+    const slidePx = 80 * p
+    if (tType === 'slideLeft') parts.push(`translateX(${-slidePx}px)`)
+    if (tType === 'slideRight') parts.push(`translateX(${slidePx}px)`)
+    if (tType === 'slideUp') parts.push(`translateY(${-slidePx}px)`)
+    if (tType === 'slideDown') parts.push(`translateY(${slidePx}px)`)
+
+    if (tType === 'zoomOut') parts.push(`scale(${1 - 0.10 * p})`)
+    if (tType === 'zoomIn') parts.push(`scale(${1 + 0.10 * p})`)
+    if (tType === 'rotateOut') parts.push(`rotate(${0.12 * p}rad)`)
+
+    return parts.length ? parts.join(' ') : 'none'
+  })()
+  
+  return (
+    <div
+      style={{
+        width: '100%',
+        height: '100%',
+        backgroundColor: bgColor,
+        opacity: opacity * outOpacity,
+        transform: outTransform,
+        filter: outFilter,
+      }}
+    >
+      <Video
+        src={clip.sourceUrl}
+        startFrom={Math.round((clip.trimStart || 0) * fps)}
+        volume={typeof volume === 'number' ? volume : 1}
+        style={{
+          width: '100%',
+          height: '100%',
+          objectFit: 'cover',
+        }}
+      />
+    </div>
+  )
+}
+
+// 字幕组件
+const SubtitleClip: React.FC<{
+  clip: TimelineClip
+  ui?: UiBridge
+  editing?: { clipId: string | null; draft: string }
+  setEditing?: React.Dispatch<React.SetStateAction<{ clipId: string | null; draft: string }>>
+}> = ({ clip, ui, editing, setEditing }) => {
+  if (!clip.text) return null
+  
+  const frame = useCurrentFrame()
+  const { fps, width, height } = useVideoConfig()
+  
+  // 淡入淡出效果
+  const opacity = interpolate(
+    frame,
+    [0, 10, clip.duration * fps - 10, clip.duration * fps],
+    [0, 1, 1, 0],
+    { extrapolateLeft: 'clamp', extrapolateRight: 'clamp' }
+  )
+  
+  const style: any = clip.style || {}
+  const isSelected = !!ui?.selectedClipId && ui.selectedClipId === clip.id
+  const isEditing = !!editing?.clipId && editing.clipId === clip.id
+
+  const boxW = (() => {
+    const bw = Number(style.box_w ?? style.boxW ?? style.max_width ?? style.maxWidth ?? 0) || 0
+    if (bw > 0 && bw <= 1) return Math.round(width * bw)
+    if (bw > 1) return Math.round(bw)
+    return Math.round(width * 0.8)
+  })()
+
+  const baseX = normPosPct(clip.position?.x, width, 0.5)
+  const baseY = normPosPct(clip.position?.y, height, (height - 220) / Math.max(1, height))
+  const renderSize = React.useContext(RenderSizeCtx)
+
+  const dragRef = React.useRef<null | { pointerId: number; sx: number; sy: number; bx: number; by: number; started: boolean }>(null)
+  const rafRef = React.useRef<number | null>(null)
+  const widthRef = React.useRef<null | { pointerId: number; sx: number; base: number; started: boolean }>(null)
+
+  const saveEdit = () => {
+    if (!isEditing) return
+    const txt = String(editing?.draft || '').replace(/\u00A0/g, ' ')
+    // 空文本：不保存，避免“看起来像被删掉”
+    if (txt.trim() === '') {
+      setEditing?.({ clipId: null, draft: '' })
+      return
+    }
+    ui?.onPushHistory?.({ label: '编辑文字', icon: 'text' })
+    ui?.onUpdateClip?.(clip.id, { text: txt })
+    setEditing?.({ clipId: null, draft: '' })
+  }
+  const cancelEdit = () => setEditing?.({ clipId: null, draft: '' })
+
+  const startDrag = (e: React.PointerEvent) => {
+    if (!ui?.onUpdateClip) return
+    if (isEditing) return
+    // 不要在 pointerdown 就 preventDefault：会破坏 click/dblclick，导致“无法双击编辑”
+    e.stopPropagation()
+    ui?.onSelectClip?.(clip.id)
+    dragRef.current = { pointerId: e.pointerId, sx: e.clientX, sy: e.clientY, bx: baseX, by: baseY, started: false }
+    const prevSel = document.body.style.userSelect
+    const prevCursor = document.body.style.cursor
+    const denomW = Math.max(1, Number(renderSize.w || 0) || 0) || width
+    const denomH = Math.max(1, Number(renderSize.h || 0) || 0) || height
+
+    const move = (ev: PointerEvent) => {
+      const d = dragRef.current
+      if (!d || d.pointerId !== ev.pointerId) return
+      const dxPx = ev.clientX - d.sx
+      const dyPx = ev.clientY - d.sy
+      if (!d.started) {
+        // 降低阈值：让拖拽更“跟手”（避免松开才明显位移的乏力感）
+        if (Math.abs(dxPx) + Math.abs(dyPx) < 1) return
+        d.started = true
+        ui?.onPushHistory?.({ label: '移动文字', icon: 'text' })
+        document.body.style.userSelect = 'none'
+        document.body.style.cursor = 'grabbing'
+      }
+      ev.preventDefault?.()
+      const nx = Math.max(0, Math.min(1, d.bx + dxPx / denomW))
+      const ny = Math.max(0, Math.min(1, d.by + dyPx / denomH))
+      if (rafRef.current) cancelAnimationFrame(rafRef.current)
+      rafRef.current = requestAnimationFrame(() => {
+        ui?.onUpdateClip?.(clip.id, { position: { x: nx, y: ny } })
+        rafRef.current = null
+      })
+    }
+    const up = (ev: PointerEvent) => {
+      const d = dragRef.current
+      if (!d || d.pointerId !== ev.pointerId) return
+      dragRef.current = null
+      try { window.removeEventListener('pointermove', move as any) } catch {}
+      try { window.removeEventListener('pointerup', up as any) } catch {}
+      try { window.removeEventListener('pointercancel', up as any) } catch {}
+      document.body.style.userSelect = prevSel
+      document.body.style.cursor = prevCursor
+    }
+    window.addEventListener('pointermove', move as any, { passive: false } as any)
+    window.addEventListener('pointerup', up as any)
+    window.addEventListener('pointercancel', up as any)
+  }
+
+  const startResizeW = (e: React.PointerEvent) => {
+    if (!ui?.onUpdateClip) return
+    if (isEditing) return
+    e.stopPropagation()
+    e.preventDefault()
+    ui?.onSelectClip?.(clip.id)
+    const base = getBoxWPct(style, width, 0.8)
+    widthRef.current = { pointerId: e.pointerId, sx: e.clientX, base, started: false }
+    const prevSel = document.body.style.userSelect
+    const prevCursor = document.body.style.cursor
+    const denomW = Math.max(1, Number(renderSize.w || 0) || 0) || width
+    const move = (ev: PointerEvent) => {
+      const d = widthRef.current
+      if (!d || d.pointerId !== ev.pointerId) return
+      const dxPx = ev.clientX - d.sx
+      if (!d.started) {
+        if (Math.abs(dxPx) < 1) return
+        d.started = true
+        ui?.onPushHistory?.({ label: '调整文字宽度', icon: 'text' })
+        document.body.style.userSelect = 'none'
+        document.body.style.cursor = 'ew-resize'
+      }
+      ev.preventDefault?.()
+      const next = clamp(d.base + dxPx / denomW, 0.2, 1.0)
+      ui?.onUpdateClip?.(clip.id, { style: { ...(style || {}), box_w: next } })
+    }
+    const up = (ev: PointerEvent) => {
+      const d = widthRef.current
+      if (!d || d.pointerId !== ev.pointerId) return
+      widthRef.current = null
+      try { window.removeEventListener('pointermove', move as any) } catch {}
+      try { window.removeEventListener('pointerup', up as any) } catch {}
+      try { window.removeEventListener('pointercancel', up as any) } catch {}
+      document.body.style.userSelect = prevSel
+      document.body.style.cursor = prevCursor
+    }
+    window.addEventListener('pointermove', move as any, { passive: false } as any)
+    window.addEventListener('pointerup', up as any)
+    window.addEventListener('pointercancel', up as any)
+  }
+
+  const enterEdit = () => {
+    ui?.onSelectClip?.(clip.id)
+    setEditing?.({ clipId: clip.id, draft: String(clip.text || '') })
+  }
+
+  return (
+    <div
+      style={{
+        position: 'absolute',
+        left: resolvePos(clip.position?.x, width, width * 0.5),
+        top: resolvePos(clip.position?.y, height, height - 220),
+        transform: 'translate(-50%, 0)',
+        width: boxW,
+        opacity,
+        pointerEvents: 'auto',
+        touchAction: 'none',
+      }}
+      onPointerDown={startDrag}
+      onClick={(e) => { e.stopPropagation(); ui?.onSelectClip?.(clip.id) }}
+      onDoubleClick={(e) => { e.preventDefault(); e.stopPropagation(); enterEdit() }}
+    >
+      <span
+        style={{
+          fontSize: (style.font_size as number) || (style.fontsize as number) || 60,
+          fontWeight: style.bold ? 800 : 600,
+          fontStyle: style.italic ? 'italic' : 'normal',
+          textDecoration: style.underline ? 'underline' : 'none',
+          fontFamily: toCssFontFamily(style.font_family),
+          color: style.font_color || '#FFFFFF',
+          textShadow: style._preview_heavy === false
+            ? 'none'
+            : '2px 2px 4px rgba(0,0,0,0.8), -2px -2px 4px rgba(0,0,0,0.8)',
+          padding: '8px 16px',
+          display: 'inline-block',
+          width: '100%',
+          whiteSpace: 'pre-wrap',
+          wordBreak: 'break-word',
+          textAlign: 'center',
+          outline: isSelected ? '1px solid rgba(88, 166, 255, 0.9)' : 'none',
+          borderRadius: 8,
+          cursor: isEditing ? 'text' : 'move',
+          userSelect: isEditing ? 'text' : 'none',
+        }}
+      >
+        {isEditing ? (
+          <span
+            contentEditable
+            suppressContentEditableWarning
+            spellCheck={false}
+            onInput={(e) => {
+              const txt = (e.currentTarget.textContent ?? '').replace(/\u00A0/g, ' ')
+              setEditing?.({ clipId: clip.id, draft: txt })
+            }}
+            onBlur={saveEdit}
+            onKeyDown={(e) => {
+              if (e.key === 'Escape') { e.preventDefault(); e.stopPropagation(); cancelEdit(); return }
+              if ((e.metaKey || e.ctrlKey) && e.key === 'Enter') { e.preventDefault(); e.stopPropagation(); saveEdit(); return }
+            }}
+            style={{ outline: 'none', display: 'inline-block', width: '100%' }}
+          >
+            {editing?.draft ?? ''}
+          </span>
+        ) : (
+          clip.text
+        )}
+      </span>
+      {/* 仅选中且非编辑：显示“拉宽”手柄（控制换行 box_w） */}
+      {isSelected && !isEditing && (
+        <div
+          style={{
+            position: 'absolute',
+            right: -6,
+            top: '50%',
+            transform: 'translateY(-50%)',
+            width: 14,
+            height: 14,
+            borderRadius: 4,
+            background: 'rgba(88,166,255,0.95)',
+            border: '1px solid rgba(255,255,255,0.65)',
+            cursor: 'ew-resize',
+          }}
+          onPointerDown={startResizeW}
+          title="拖动拉宽（控制换行）"
+        />
+      )}
+    </div>
+  )
+}
+
+// 花字组件
+const FancyTextClip: React.FC<{
+  clip: TimelineClip
+  ui?: UiBridge
+  editing?: { clipId: string | null; draft: string }
+  setEditing?: React.Dispatch<React.SetStateAction<{ clipId: string | null; draft: string }>>
+}> = ({ clip, ui, editing, setEditing }) => {
+  if (!clip.text) return null
+  
+  const frame = useCurrentFrame()
+  const { width, height } = useVideoConfig()
+  
+  // 弹入效果
+  const scale = interpolate(
+    frame,
+    [0, 15],
+    [0.5, 1],
+    { extrapolateLeft: 'clamp', extrapolateRight: 'clamp' }
+  )
+  
+  const style: any = clip.style || {}
+  const isSelected = !!ui?.selectedClipId && ui.selectedClipId === clip.id
+  const isEditing = !!editing?.clipId && editing.clipId === clip.id
+  const position = clip.position || { x: 0.5, y: 0.2 }
+  const x = resolvePos((position as any).x, width, width * 0.5)
+  const y = resolvePos((position as any).y, height, 180)
+  const baseX = normPosPct((position as any).x, width, 0.5)
+  const baseY = normPosPct((position as any).y, height, 180 / Math.max(1, height))
+  const renderSize = React.useContext(RenderSizeCtx)
+
+  const dragRef = React.useRef<null | { pointerId: number; sx: number; sy: number; bx: number; by: number; started: boolean }>(null)
+  const rafRef = React.useRef<number | null>(null)
+  const widthRef = React.useRef<null | { pointerId: number; sx: number; base: number; started: boolean }>(null)
+
+  const saveEdit = () => {
+    if (!isEditing) return
+    const txt = String(editing?.draft || '').replace(/\u00A0/g, ' ')
+    // 空文本：不保存，避免“看起来像被删掉”
+    if (txt.trim() === '') {
+      setEditing?.({ clipId: null, draft: '' })
+      return
+    }
+    ui?.onPushHistory?.({ label: '编辑文字', icon: 'text' })
+    ui?.onUpdateClip?.(clip.id, { text: txt })
+    setEditing?.({ clipId: null, draft: '' })
+  }
+  const cancelEdit = () => setEditing?.({ clipId: null, draft: '' })
+
+  const startDrag = (e: React.PointerEvent) => {
+    if (!ui?.onUpdateClip) return
+    if (isEditing) return
+    // 不要在 pointerdown 就 preventDefault：会破坏 click/dblclick，导致“无法双击编辑”
+    e.stopPropagation()
+    ui?.onSelectClip?.(clip.id)
+    dragRef.current = { pointerId: e.pointerId, sx: e.clientX, sy: e.clientY, bx: baseX, by: baseY, started: false }
+    const prevSel = document.body.style.userSelect
+    const prevCursor = document.body.style.cursor
+    const denomW = Math.max(1, Number(renderSize.w || 0) || 0) || width
+    const denomH = Math.max(1, Number(renderSize.h || 0) || 0) || height
+
+    const move = (ev: PointerEvent) => {
+      const d = dragRef.current
+      if (!d || d.pointerId !== ev.pointerId) return
+      const dxPx = ev.clientX - d.sx
+      const dyPx = ev.clientY - d.sy
+      if (!d.started) {
+        // 降低阈值：让拖拽更“跟手”（避免松开才明显位移的乏力感）
+        if (Math.abs(dxPx) + Math.abs(dyPx) < 1) return
+        d.started = true
+        ui?.onPushHistory?.({ label: '移动文字', icon: 'text' })
+        document.body.style.userSelect = 'none'
+        document.body.style.cursor = 'grabbing'
+      }
+      ev.preventDefault?.()
+      const nx = Math.max(0, Math.min(1, d.bx + dxPx / denomW))
+      const ny = Math.max(0, Math.min(1, d.by + dyPx / denomH))
+      if (rafRef.current) cancelAnimationFrame(rafRef.current)
+      rafRef.current = requestAnimationFrame(() => {
+        ui?.onUpdateClip?.(clip.id, { position: { x: nx, y: ny } })
+        rafRef.current = null
+      })
+    }
+    const up = (ev: PointerEvent) => {
+      const d = dragRef.current
+      if (!d || d.pointerId !== ev.pointerId) return
+      dragRef.current = null
+      try { window.removeEventListener('pointermove', move as any) } catch {}
+      try { window.removeEventListener('pointerup', up as any) } catch {}
+      try { window.removeEventListener('pointercancel', up as any) } catch {}
+      document.body.style.userSelect = prevSel
+      document.body.style.cursor = prevCursor
+    }
+    window.addEventListener('pointermove', move as any, { passive: false } as any)
+    window.addEventListener('pointerup', up as any)
+    window.addEventListener('pointercancel', up as any)
+  }
+
+  const startResizeW = (e: React.PointerEvent) => {
+    if (!ui?.onUpdateClip) return
+    if (isEditing) return
+    e.stopPropagation()
+    e.preventDefault()
+    ui?.onSelectClip?.(clip.id)
+    const base = getBoxWPct(style, width, 0.7)
+    widthRef.current = { pointerId: e.pointerId, sx: e.clientX, base, started: false }
+    const prevSel = document.body.style.userSelect
+    const prevCursor = document.body.style.cursor
+    const denomW = Math.max(1, Number(renderSize.w || 0) || 0) || width
+    const move = (ev: PointerEvent) => {
+      const d = widthRef.current
+      if (!d || d.pointerId !== ev.pointerId) return
+      const dxPx = ev.clientX - d.sx
+      if (!d.started) {
+        if (Math.abs(dxPx) < 1) return
+        d.started = true
+        ui?.onPushHistory?.({ label: '调整文字宽度', icon: 'text' })
+        document.body.style.userSelect = 'none'
+        document.body.style.cursor = 'ew-resize'
+      }
+      ev.preventDefault?.()
+      const next = clamp(d.base + dxPx / denomW, 0.2, 1.0)
+      ui?.onUpdateClip?.(clip.id, { style: { ...(style || {}), box_w: next } })
+    }
+    const up = (ev: PointerEvent) => {
+      const d = widthRef.current
+      if (!d || d.pointerId !== ev.pointerId) return
+      widthRef.current = null
+      try { window.removeEventListener('pointermove', move as any) } catch {}
+      try { window.removeEventListener('pointerup', up as any) } catch {}
+      try { window.removeEventListener('pointercancel', up as any) } catch {}
+      document.body.style.userSelect = prevSel
+      document.body.style.cursor = prevCursor
+    }
+    window.addEventListener('pointermove', move as any, { passive: false } as any)
+    window.addEventListener('pointerup', up as any)
+    window.addEventListener('pointercancel', up as any)
+  }
+
+  const enterEdit = () => {
+    ui?.onSelectClip?.(clip.id)
+    setEditing?.({ clipId: clip.id, draft: String(clip.text || '') })
+  }
+  
+  return (
+    <div
+      style={{
+        position: 'absolute',
+        left: x,
+        top: y,
+        transform: `translate(-50%, 0) scale(${scale})`,
+        pointerEvents: 'auto',
+        touchAction: 'none',
+      }}
+      onPointerDown={startDrag}
+      onClick={(e) => { e.stopPropagation(); ui?.onSelectClip?.(clip.id) }}
+      onDoubleClick={(e) => { e.preventDefault(); e.stopPropagation(); enterEdit() }}
+    >
+      <span
+        style={{
+          fontSize: (style.font_size as number) || 72,
+          fontWeight: (style as any).bold ? 800 : 700,
+          fontStyle: (style as any).italic ? 'italic' : 'normal',
+          textDecoration: (style as any).underline ? 'underline' : 'none',
+          fontFamily: toCssFontFamily((style as any).font_family),
+          color: (style.font_color as string) || '#FFFFFF',
+          textShadow: '3px 3px 6px rgba(0,0,0,0.9), -3px -3px 6px rgba(0,0,0,0.9)',
+          display: 'inline-block',
+          maxWidth: (() => {
+            const bw = Number((style as any).box_w ?? (style as any).boxW ?? (style as any).max_width ?? (style as any).maxWidth ?? 0) || 0
+            if (bw > 0 && bw <= 1) return Math.round(width * bw)
+            if (bw > 1) return Math.round(bw)
+            return Math.round(width * 0.7)
+          })(),
+          whiteSpace: 'pre-wrap',
+          wordBreak: 'break-word',
+          textAlign: 'center',
+          outline: isSelected ? '1px solid rgba(88, 166, 255, 0.9)' : 'none',
+          borderRadius: 10,
+          cursor: isEditing ? 'text' : 'move',
+          userSelect: isEditing ? 'text' : 'none',
+        }}
+      >
+        {isEditing ? (
+          <span
+            contentEditable
+            suppressContentEditableWarning
+            spellCheck={false}
+            onInput={(e) => {
+              const txt = (e.currentTarget.textContent ?? '').replace(/\u00A0/g, ' ')
+              setEditing?.({ clipId: clip.id, draft: txt })
+            }}
+            onBlur={saveEdit}
+            onKeyDown={(e) => {
+              if (e.key === 'Escape') { e.preventDefault(); e.stopPropagation(); cancelEdit(); return }
+              if ((e.metaKey || e.ctrlKey) && e.key === 'Enter') { e.preventDefault(); e.stopPropagation(); saveEdit(); return }
+            }}
+            style={{ outline: 'none', display: 'inline-block' }}
+          >
+            {editing?.draft ?? ''}
+          </span>
+        ) : (
+          clip.text
+        )}
+      </span>
+      {/* 仅选中且非编辑：显示“拉宽”手柄（控制换行 box_w） */}
+      {isSelected && !isEditing && (
+        <div
+          style={{
+            position: 'absolute',
+            right: -6,
+            top: '50%',
+            transform: 'translateY(-50%)',
+            width: 14,
+            height: 14,
+            borderRadius: 4,
+            background: 'rgba(88,166,255,0.95)',
+            border: '1px solid rgba(255,255,255,0.65)',
+            cursor: 'ew-resize',
+          }}
+          onPointerDown={startResizeW}
+          title="拖动拉宽（控制换行）"
+        />
+      )}
+    </div>
+  )
+}
+
+// 贴纸组件
+const StickerClip: React.FC<{
+  clip: TimelineClip
+}> = ({ clip }) => {
+  if (!clip.sourceUrl) return null
+  const { width, height } = useVideoConfig()
+  const style: any = clip.style || {}
+  const position = clip.position || { x: 0.8, y: 0.2 }
+  const x = resolvePos((position as any).x, width, width * 0.8)
+  const y = resolvePos((position as any).y, height, height * 0.2)
+  const scale = Math.max(0.3, Math.min(3.0, Number(style.scale ?? 1) || 1))
+  const rotate = Number(style.rotate ?? 0) || 0
+  const base = Math.round(Math.min(width, height) * 0.22)
+  return (
+    <div
+      style={{
+        position: 'absolute',
+        left: x,
+        top: y,
+        transform: `translate(-50%, -50%) scale(${scale}) rotate(${rotate}deg)`,
+        transformOrigin: 'center',
+        width: base,
+        height: base,
+        pointerEvents: 'none',
+      }}
+    >
+      <Img src={clip.sourceUrl} style={{ width: '100%', height: '100%', objectFit: 'contain' }} />
+    </div>
+  )
+}
+
+// 音频组件
+const AudioClip: React.FC<{
+  clip: TimelineClip
+  volume?: (frame: number) => number
+}> = ({ clip, volume }) => {
+  if (!clip.sourceUrl) return null
+  
+  return (
+    <Audio
+      src={clip.sourceUrl}
+      volume={volume ?? (clip.volume || 1)}
+      playbackRate={Number((clip as any).playbackRate ?? 1) || 1}
+      loop={Boolean((clip as any).loop ?? (clip as any).style?.loop)}
+    />
+  )
+}
+
+// Props
+export interface VideoCompositionProps {
+  tracks: Track[]
+  fps?: number
+  subtitleStyle?: Record<string, unknown>
+  preview?: {
+    heavyOverlays?: boolean
+  }
+  audio?: {
+    soloTrackIds?: string[]
+  }
+  ui?: UiBridge
+}
+
+// 主合成组件
+export const VideoComposition: React.FC<VideoCompositionProps> = ({
+  tracks,
+  subtitleStyle,
+  preview,
+  audio,
+  ui,
+}) => {
+  const { fps } = useVideoConfig()
+  const heavy = preview?.heavyOverlays !== false
+  const soloSet = new Set((audio?.soloTrackIds || []).filter(Boolean))
+  const hasSolo = soloSet.size > 0
+
+  const [editing, setEditing] = React.useState<{ clipId: string | null; draft: string }>({ clipId: null, draft: '' })
+  const [renderSize, setRenderSize] = React.useState<{ w: number; h: number }>({ w: 0, h: 0 })
+  const sizeProbeRef = React.useRef<HTMLDivElement | null>(null)
+
+  React.useEffect(() => {
+    const el = sizeProbeRef.current
+    if (!el) return
+    const update = () => {
+      const r = el.getBoundingClientRect()
+      const w = Math.round(r.width || 0)
+      const h = Math.round(r.height || 0)
+      if (w > 0 && h > 0) setRenderSize({ w, h })
+    }
+    update()
+    const ro = new ResizeObserver(() => update())
+    try { ro.observe(el) } catch {}
+    const id = window.setInterval(update, 500) // 兜底：部分浏览器 ResizeObserver 偶发不触发
+    return () => {
+      try { ro.disconnect() } catch {}
+      clearInterval(id)
+    }
+  }, [])
+  
+  // 分离不同类型的轨道
+  const videoTrack = tracks.find(t => t.type === 'video')
+  const subtitleTrack = tracks.find(t => t.type === 'subtitle')
+  const fancyTextTrack = tracks.find(t => t.type === 'fancy_text')
+  const stickerTrack = tracks.find(t => t.type === 'sticker')
+  const voiceoverTrack = tracks.find(t => t.id === 'audio-voiceover')
+  const bgmTrack = tracks.find(t => t.type === 'bgm')
+
+  // 预计算旁白区间（用于 BGM ducking）
+  const voRanges = (voiceoverTrack?.clips || []).map(c => {
+    const s = c.start ?? 0
+    const e = s + (c.duration ?? 0)
+    return { s, e }
+  })
+  const isVoiceAt = (t: number) => voRanges.some(r => t >= r.s && t <= r.e)
+  
+  return (
+    <RenderSizeCtx.Provider value={renderSize}>
+      <AbsoluteFill style={{ backgroundColor: '#000' }}>
+        {/* 仅用于测量“实际渲染像素尺寸”（用于拖拽严格跟随鼠标） */}
+        <div ref={sizeProbeRef} style={{ position: 'absolute', inset: 0, pointerEvents: 'none' }} />
+      {/* 视频层 */}
+      {videoTrack?.clips.map(clip => (
+        <Sequence
+          key={clip.id}
+          from={(() => {
+            // 用 floor/ceil 避免浮点取整导致的“1帧黑屏缝隙”
+            const s = Math.max(0, clip.start ?? 0)
+            return Math.floor(s * fps)
+          })()}
+          durationInFrames={(() => {
+            const s = Math.max(0, clip.start ?? 0)
+            const e = s + Math.max(0, clip.duration ?? 0)
+            const startF = Math.floor(s * fps)
+            const endF = Math.max(startF + 1, Math.ceil(e * fps))
+            return endF - startF
+          })()}
+        >
+          <VideoClip
+            clip={clip}
+            volume={
+              (videoTrack && (!hasSolo || soloSet.has(videoTrack.id)) && videoTrack.muted !== true)
+                ? ((clip.volume ?? 0.2) as number)
+                : 0
+            }
+          />
+        </Sequence>
+      ))}
+
+      {/* 贴纸层（在文字下方，避免挡字幕；可后续支持层级调整） */}
+      {stickerTrack?.visible !== false && stickerTrack?.clips?.map(clip => (
+        <Sequence
+          key={clip.id}
+          from={Math.round((clip.start ?? 0) * fps)}
+          durationInFrames={Math.round((clip.duration ?? 0) * fps)}
+        >
+          <StickerClip clip={clip} />
+        </Sequence>
+      ))}
+      
+      {/* 花字层 */}
+      {fancyTextTrack?.visible !== false && fancyTextTrack?.clips.map(clip => (
+        <Sequence
+          key={clip.id}
+          from={Math.round(clip.start * fps)}
+          durationInFrames={Math.round(clip.duration * fps)}
+        >
+          <FancyTextClip
+            clip={{ ...clip, style: { ...(clip.style || {}), _preview_heavy: heavy } }}
+            ui={ui}
+            editing={editing}
+            setEditing={setEditing}
+          />
+        </Sequence>
+      ))}
+      
+      {/* 字幕层 */}
+      {subtitleTrack?.visible !== false && subtitleTrack?.clips.map(clip => (
+        <Sequence
+          key={clip.id}
+          from={Math.round(clip.start * fps)}
+          durationInFrames={Math.round(clip.duration * fps)}
+        >
+          <SubtitleClip
+            clip={{ ...clip, style: { ...(subtitleStyle || {}), ...(clip.style || {}), _preview_heavy: heavy } }}
+            ui={ui}
+            editing={editing}
+            setEditing={setEditing}
+          />
+        </Sequence>
+      ))}
+      
+      {/* 旁白音频 */}
+      {(voiceoverTrack && (!hasSolo || soloSet.has(voiceoverTrack.id)) && voiceoverTrack?.muted !== true) && voiceoverTrack.clips.map(clip => (
+        <Sequence
+          key={clip.id}
+          from={Math.round(clip.start * fps)}
+          durationInFrames={Math.round(clip.duration * fps)}
+        >
+          <AudioClip
+            clip={clip}
+            volume={(frame) => {
+              const from = Math.round((clip.start ?? 0) * fps)
+              const durF = Math.max(1, Math.round((clip.duration ?? 0) * fps))
+              const local = frame - from
+              const fade = makeFade(local, durF, fps, clip.fadeIn, clip.fadeOut)
+              return clamp01((clip.volume ?? 1) * fade)
+            }}
+          />
+        </Sequence>
+      ))}
+      
+      {/* BGM */}
+      {(bgmTrack && (!hasSolo || soloSet.has(bgmTrack.id)) && bgmTrack?.muted !== true) && bgmTrack.clips.map(clip => (
+        <Sequence
+          key={clip.id}
+          from={Math.round(clip.start * fps)}
+          durationInFrames={Math.round(clip.duration * fps)}
+        >
+          <AudioClip
+            clip={clip}
+            volume={(frame) => {
+              const from = Math.round((clip.start ?? 0) * fps)
+              const durF = Math.max(1, Math.round((clip.duration ?? 0) * fps))
+              const local = frame - from
+              const t = frame / fps
+              const fade = makeFade(local, durF, fps, clip.fadeIn ?? 0.8, clip.fadeOut ?? 0.8)
+              const base = (clip.volume ?? 0.15) as number
+              const duckEnabled = typeof clip.ducking === 'boolean' ? clip.ducking : true
+              const duckVol = (clip.duckVolume ?? 0.25) as number
+              const duck = duckEnabled && isVoiceAt(t) ? duckVol : 1
+              return clamp01(base * fade * duck)
+            }}
+          />
+        </Sequence>
+      ))}
+      </AbsoluteFill>
+    </RenderSizeCtx.Provider>
+  )
+}
+
+export default VideoComposition
+

web/src/remotion/index.ts

@@ -0,0 +1,18 @@
+/**
+ * Remotion 组件导出
+ */
+export { VideoComposition } from './VideoComposition'
+
+
+
+
+
+
+
+
+
+
+
+
+
+

web/src/store/editorStore.ts

@@ -0,0 +1,734 @@
+/**
+ * 编辑器状态管理 (Zustand)
+ * 管理时间轴、轨道、播放状态
+ */
+import { create } from 'zustand'
+import { immer } from 'zustand/middleware/immer'
+import { generateId } from '@/lib/utils'
+
+// 类型定义
+export interface TimelineClip {
+  id: string
+  type: 'video' | 'audio' | 'subtitle' | 'fancy_text' | 'bgm' | 'sticker'
+  start: number
+  duration: number
+  sourcePath?: string
+  sourceUrl?: string
+  trimStart?: number
+  trimEnd?: number
+  sourceDuration?: number
+  text?: string
+  style?: Record<string, unknown>
+  position?: { x: string | number; y: string | number }
+  volume?: number
+  fadeIn?: number
+  fadeOut?: number
+  // 仅 BGM：自动闪避（有人声时降低到 duckVolume 倍）
+  ducking?: boolean
+  duckVolume?: number
+  // 旁白：纯播放倍速（预览=导出）。1=正常，0.5=慢，2=快
+  playbackRate?: number
+  groupId?: string
+  /**
+   * 仅旁白：当用户修改了文本/时长后，现有配音可能不再匹配。
+   * 用于 UX 提示（不参与导出逻辑、也不需要持久化）。
+   */
+  needsVoiceoverRegenerate?: boolean
+}
+
+export interface Track {
+  id: string
+  name: string
+  type: string
+  clips: TimelineClip[]
+  locked: boolean
+  visible: boolean
+  muted: boolean
+  collapsed?: boolean
+}
+
+export interface HistoryEntry {
+  tracks: Track[]
+  label: string
+  icon?: string
+  ts: number
+}
+
+interface EditorState {
+  // 项目信息
+  projectId: string | null
+  
+  // 时间轴状态
+  tracks: Track[]
+  totalDuration: number
+  currentTime: number
+  isPlaying: boolean
+
+  // 编辑策略
+  rippleMode: boolean
+
+  // 音频：轨道 Solo（任意 Solo 时，仅 Solo 轨道参与预览播放）
+  soloTrackIds: string[]
+
+  // 字幕全局样式（MVP）
+  subtitleStyle: Record<string, unknown>
+
+  // 预览性能档位（不影响最终导出）
+  previewFps: number
+  previewScale: number
+  previewHeavyOverlays: boolean
+  
+  // 缩放和滚动
+  zoom: number
+  scrollX: number
+  
+  // 选中状态
+  selectedClipId: string | null
+  selectedTrackId: string | null
+  selectedClipIds: string[]
+
+  // 辅助线（吸附/对齐提示）
+  snapGuideTime: number | null
+  
+  // 撤销/重做栈（最多 30 步）
+  history: HistoryEntry[]
+  historyIndex: number
+  
+  // 加载状态
+  isLoading: boolean
+  isSaving: boolean
+  
+  // Actions
+  setProjectId: (id: string) => void
+  setTracks: (tracks: Track[]) => void
+  setTotalDuration: (duration: number) => void
+  setCurrentTime: (time: number) => void
+  setIsPlaying: (playing: boolean) => void
+  setRippleMode: (enabled: boolean) => void
+  toggleTrackSolo: (trackId: string) => void
+  setSubtitleStyle: (style: Record<string, unknown>) => void
+  setPreviewQuality: (q: Partial<Pick<EditorState, 'previewFps' | 'previewScale' | 'previewHeavyOverlays'>>) => void
+  setZoom: (zoom: number) => void
+  setScrollX: (x: number) => void
+  
+  // 选择
+  selectClip: (clipId: string | null) => void
+  selectTrack: (trackId: string | null) => void
+  setSelectedClips: (clipIds: string[]) => void
+  toggleClipSelection: (clipId: string, mode?: 'replace' | 'toggle' | 'add') => void
+  clearSelection: () => void
+  setSnapGuideTime: (t: number | null) => void
+
+  // 分组（跨轨联动）
+  groupSelected: (groupId?: string) => void
+  ungroupSelected: () => void
+  shiftGroup: (groupId: string, delta: number) => void
+  
+  // 片段操作
+  addClip: (trackId: string, clip: Omit<TimelineClip, 'id'>) => void
+  updateClip: (trackId: string, clipId: string, updates: Partial<TimelineClip>) => void
+  deleteClip: (trackId: string, clipId: string) => void
+  deleteSelectedClips: () => void
+  deleteAtPlayhead: () => void
+  moveClip: (fromTrackId: string, toTrackId: string, clipId: string, newStart: number) => void
+  splitClip: (trackId: string, clipId: string, time: number) => void
+  splitAtTime: (time: number) => void
+  rippleShiftFrom: (trackId: string, clipId: string, delta: number) => void
+  
+  // 轨道操作
+  addTrack: (track: Omit<Track, 'id'>) => void
+  updateTrack: (trackId: string, updates: Partial<Track>) => void
+  deleteTrack: (trackId: string) => void
+  toggleTrackMute: (trackId: string) => void
+  toggleTrackLock: (trackId: string) => void
+  toggleTrackCollapse: (trackId: string) => void
+  
+  // 历史操作
+  pushHistory: (meta?: { label?: string; icon?: string }) => void
+  undo: () => void
+  redo: () => void
+  jumpToHistory: (index: number) => void
+  
+  // 重置
+  reset: () => void
+}
+
+const initialState = {
+  projectId: null,
+  tracks: [],
+  totalDuration: 0,
+  currentTime: 0,
+  isPlaying: false,
+  rippleMode: true,
+  soloTrackIds: [],
+  subtitleStyle: {
+    fontsize: 60,
+    fontcolor: 'white',
+    borderw: 5,
+    bordercolor: 'black',
+    box: 1,
+    boxcolor: 'black@0.5',
+    boxborderw: 10,
+    x: '(w-text_w)/2',
+    y: 'h-200',
+  },
+  previewFps: 30,
+  previewScale: 1,
+  previewHeavyOverlays: true,
+  zoom: 1,
+  scrollX: 0,
+  selectedClipId: null,
+  selectedTrackId: null,
+  selectedClipIds: [],
+  snapGuideTime: null,
+  history: [],
+  historyIndex: -1,
+  isLoading: false,
+  isSaving: false,
+}
+
+export const useEditorStore = create<EditorState>()(
+  immer((set) => {
+    const recalcTotalDuration = (tracks: Track[]) => {
+      let maxEnd = 0
+      for (const t of tracks || []) {
+        for (const c of t.clips || []) {
+          const end = (c.start || 0) + (c.duration || 0)
+          if (end > maxEnd) maxEnd = end
+        }
+      }
+      return maxEnd
+    }
+
+    const clampVideoDur = (c: TimelineClip, minDur: number) => {
+      let trimStart = c.trimStart ?? 0
+      let duration = Math.max(minDur, c.duration ?? minDur)
+      if (typeof c.sourceDuration === 'number' && c.sourceDuration > 0) {
+        // trimStart 不能超过 sourceDuration-minDur，否则会出现 trimEnd 超界导致预览卡住
+        trimStart = Math.max(0, Math.min(trimStart, Math.max(0, c.sourceDuration - minDur)))
+        duration = Math.min(duration, Math.max(minDur, c.sourceDuration - trimStart))
+      }
+      c.trimStart = trimStart
+      c.duration = duration
+      c.trimEnd = trimStart + duration
+    }
+
+    // 兜底：强制同轨道视频永不重叠（默认 Ripple：发生碰撞时推挤后续片段）
+    // 规则：
+    // - 不允许与前一个片段重叠：当前片段会被 clamp 到 prevEnd
+    // - 不允许与后一个片段重叠：后续片段会被整体 push 到不重叠（Ripple）
+    const enforceVideoNoOverlap = (track: Track) => {
+      if (!track || track.type !== 'video') return
+      const minDur = 0.5
+      track.clips.sort((a, b) => (a.start ?? 0) - (b.start ?? 0))
+
+      // 基础数据修正 + sourceDuration 保护
+      for (const c of track.clips) {
+        if (typeof c.start !== 'number' || !Number.isFinite(c.start)) c.start = 0
+        if (typeof c.duration !== 'number' || !Number.isFinite(c.duration)) c.duration = minDur
+        c.start = Math.max(0, c.start)
+        clampVideoDur(c, minDur)
+      }
+
+      // Ripple 推挤：确保每个 clip.start >= prevEnd
+      let prevEnd = 0
+      for (let i = 0; i < track.clips.length; i++) {
+        const c = track.clips[i]
+        const s = c.start ?? 0
+        if (s < prevEnd - 1e-6) {
+          c.start = prevEnd
+        }
+        prevEnd = (c.start ?? 0) + (c.duration ?? 0)
+      }
+    }
+
+    return ({
+    ...initialState,
+    
+    setProjectId: (id) => set({ projectId: id }),
+    
+    setTracks: (tracks) => set({
+      tracks,
+      totalDuration: recalcTotalDuration(tracks),
+    }),
+    
+    setTotalDuration: (duration) => set({ totalDuration: duration }),
+    
+    setCurrentTime: (time) => set({ currentTime: time }),
+    
+    setIsPlaying: (playing) => set({ isPlaying: playing }),
+
+    // Ripple 默认强制开启（去掉开关，避免用户困惑）
+    setRippleMode: (_enabled) => set({ rippleMode: true }),
+
+    toggleTrackSolo: (trackId) => set((state) => {
+      const cur = new Set(state.soloTrackIds || [])
+      if (cur.has(trackId)) cur.delete(trackId)
+      else cur.add(trackId)
+      state.soloTrackIds = Array.from(cur)
+    }),
+
+    setSubtitleStyle: (style) => set((state) => {
+      state.subtitleStyle = { ...(state.subtitleStyle || {}), ...(style || {}) }
+    }),
+
+    setPreviewQuality: (q) => set((state) => {
+      if (typeof q.previewFps === 'number' && q.previewFps > 0) state.previewFps = q.previewFps
+      if (typeof q.previewScale === 'number' && q.previewScale > 0) state.previewScale = q.previewScale
+      if (typeof q.previewHeavyOverlays === 'boolean') state.previewHeavyOverlays = q.previewHeavyOverlays
+    }),
+    
+    setZoom: (zoom) => set({ zoom: Math.max(0.1, Math.min(10, zoom)) }),
+    
+    setScrollX: (x) => set({ scrollX: Math.max(0, x) }),
+    
+    selectClip: (clipId) => set((state) => {
+      state.selectedClipId = clipId
+      state.selectedClipIds = clipId ? [clipId] : []
+      // 自动推断 trackId，方便快捷键/属性面板
+      if (!clipId) {
+        state.selectedTrackId = null
+        return
+      }
+      for (const t of state.tracks) {
+        if (t.clips.some(c => c.id === clipId)) {
+          state.selectedTrackId = t.id
+          return
+        }
+      }
+    }),
+    
+    selectTrack: (trackId) => set({ selectedTrackId: trackId }),
+
+    setSelectedClips: (clipIds) => set((state) => {
+      const uniq = Array.from(new Set((clipIds || []).filter(Boolean)))
+      state.selectedClipIds = uniq
+      state.selectedClipId = uniq.length ? uniq[uniq.length - 1] : null
+      // 尝试推断 selectedTrackId
+      state.selectedTrackId = null
+      if (state.selectedClipId) {
+        for (const t of state.tracks) {
+          if (t.clips.some(c => c.id === state.selectedClipId)) {
+            state.selectedTrackId = t.id
+            break
+          }
+        }
+      }
+    }),
+
+    toggleClipSelection: (clipId, mode = 'replace') => set((state) => {
+      const cur = new Set(state.selectedClipIds || [])
+      if (mode === 'replace') {
+        state.selectedClipIds = clipId ? [clipId] : []
+        state.selectedClipId = clipId || null
+      } else if (mode === 'toggle') {
+        if (cur.has(clipId)) cur.delete(clipId); else cur.add(clipId)
+        state.selectedClipIds = Array.from(cur)
+        state.selectedClipId = state.selectedClipIds.length ? state.selectedClipIds[state.selectedClipIds.length - 1] : null
+      } else {
+        cur.add(clipId)
+        state.selectedClipIds = Array.from(cur)
+        state.selectedClipId = clipId
+      }
+
+      // 推断 trackId
+      state.selectedTrackId = null
+      if (state.selectedClipId) {
+        for (const t of state.tracks) {
+          if (t.clips.some(c => c.id === state.selectedClipId)) {
+            state.selectedTrackId = t.id
+            break
+          }
+        }
+      }
+    }),
+
+    clearSelection: () => set((state) => {
+      state.selectedClipId = null
+      state.selectedTrackId = null
+      state.selectedClipIds = []
+    }),
+
+    setSnapGuideTime: (t) => set((state) => {
+      state.snapGuideTime = t
+    }),
+    
+    addClip: (trackId, clip) => set((state) => {
+      const track = state.tracks.find(t => t.id === trackId)
+      if (track && !track.locked) {
+        const newClip = { ...clip, id: generateId() }
+        track.clips.push(newClip as TimelineClip)
+        track.clips.sort((a, b) => a.start - b.start)
+        if (track.type === 'video') enforceVideoNoOverlap(track)
+        state.totalDuration = recalcTotalDuration(state.tracks)
+      }
+    }),
+    
+    updateClip: (trackId, clipId, updates) => set((state) => {
+      const track = state.tracks.find(t => t.id === trackId)
+      if (!track) return
+
+      // 轨道“锁定：防误触”主要防止误拖时间轴上的 start/duration。
+      // 但字幕/花字/贴纸的画面内位置/样式/文本属于“有意操作”，锁定也应允许，否则会出现“选中但拖不动”。
+      if (track.locked) {
+        const allowTrack = track.type === 'subtitle' || track.type === 'fancy_text' || track.type === 'sticker'
+        const keys = Object.keys(updates || {})
+        const allowKeys = keys.every(k => k === 'position' || k === 'style' || k === 'text')
+        if (!allowTrack || !allowKeys) return
+      }
+
+      {
+        const clip = track.clips.find(c => c.id === clipId)
+        if (clip) {
+          const prevDur = clip.duration
+          const prevText = clip.text
+          Object.assign(clip, updates)
+
+          // UX：旁白片段被拉伸/改文案后，提醒用户“需要重新生成以适配时长/文本”
+          if (trackId === 'audio-voiceover') {
+            const textChanged = typeof updates.text === 'string' && updates.text !== prevText
+
+            // 如果用户更新了音频来源（重新生成成功），就清除提示
+            const hasNewAudio = typeof updates.sourceUrl === 'string' || typeof updates.sourcePath === 'string'
+            if (hasNewAudio) {
+              clip.needsVoiceoverRegenerate = false
+            } else if (textChanged && (clip.text || '').trim().length > 0) {
+              clip.needsVoiceoverRegenerate = true
+            }
+
+            // 纯播放倍速：当用户改动旁白片段时长，自动计算倍速以“贴合当前时长”
+            if (typeof updates.duration === 'number' && Number.isFinite(updates.duration) && updates.duration > 0) {
+              const srcDur = typeof clip.sourceDuration === 'number' && clip.sourceDuration > 0
+                ? clip.sourceDuration
+                : (typeof prevDur === 'number' && prevDur > 0 ? prevDur : undefined)
+              if (srcDur) {
+                const r = srcDur / Math.max(0.01, clip.duration)
+                clip.playbackRate = Math.max(0.5, Math.min(2.0, Number(r) || 1))
+                // 拉长/缩短已由倍速适配，不需要“重新生成”提示
+                clip.needsVoiceoverRegenerate = false
+              }
+            }
+          }
+
+          if (track.type === 'video') {
+            enforceVideoNoOverlap(track)
+          }
+          state.totalDuration = recalcTotalDuration(state.tracks)
+        }
+      }
+    }),
+    
+    deleteClip: (trackId, clipId) => set((state) => {
+      const track = state.tracks.find(t => t.id === trackId)
+      if (track && !track.locked) {
+        track.clips = track.clips.filter(c => c.id !== clipId)
+        state.totalDuration = recalcTotalDuration(state.tracks)
+      }
+      if (state.selectedClipId === clipId) {
+        state.selectedClipId = null
+      }
+      if (state.selectedClipIds?.includes(clipId)) {
+        state.selectedClipIds = state.selectedClipIds.filter(id => id !== clipId)
+      }
+    }),
+
+    deleteSelectedClips: () => set((state) => {
+      const ids = new Set((state.selectedClipIds || []).filter(Boolean))
+      if (!ids.size) return
+      for (const t of state.tracks) {
+        if (t.locked) continue
+        const before = t.clips.length
+        t.clips = t.clips.filter(c => !ids.has(c.id))
+        if (t.type === 'video' && t.clips.length !== before) {
+          enforceVideoNoOverlap(t)
+        }
+      }
+      state.selectedClipIds = []
+      state.selectedClipId = null
+      state.selectedTrackId = null
+      state.totalDuration = recalcTotalDuration(state.tracks)
+    }),
+
+    deleteAtPlayhead: () => set((state) => {
+      const t = Math.max(0, state.currentTime ?? 0)
+      // 优先删“当前选中”
+      if (state.selectedClipId && state.selectedTrackId) {
+        const tr = state.tracks.find(x => x.id === state.selectedTrackId)
+        if (tr && !tr.locked) {
+          const exists = tr.clips.find(c => c.id === state.selectedClipId)
+          if (exists) {
+            tr.clips = tr.clips.filter(c => c.id !== state.selectedClipId)
+            if (tr.type === 'video') enforceVideoNoOverlap(tr)
+            state.selectedClipId = null
+            state.selectedTrackId = null
+            state.selectedClipIds = (state.selectedClipIds || []).filter(id => id !== exists.id)
+            state.totalDuration = recalcTotalDuration(state.tracks)
+            return
+          }
+        }
+      }
+
+      // 否则删红线所在的“视频片段”
+      const vTrack = state.tracks.find(x => x.type === 'video')
+      if (!vTrack || vTrack.locked) return
+      const clip = (vTrack.clips || []).find(c => t >= (c.start ?? 0) && t < ((c.start ?? 0) + (c.duration ?? 0)))
+      if (!clip) return
+      vTrack.clips = vTrack.clips.filter(c => c.id !== clip.id)
+      enforceVideoNoOverlap(vTrack)
+      state.totalDuration = recalcTotalDuration(state.tracks)
+      if (state.selectedClipIds?.includes(clip.id)) state.selectedClipIds = state.selectedClipIds.filter(id => id !== clip.id)
+      if (state.selectedClipId === clip.id) state.selectedClipId = null
+      if (state.selectedTrackId === vTrack.id) state.selectedTrackId = null
+    }),
+    
+    moveClip: (fromTrackId, toTrackId, clipId, newStart) => set((state) => {
+      const fromTrack = state.tracks.find(t => t.id === fromTrackId)
+      const toTrack = state.tracks.find(t => t.id === toTrackId)
+      
+      if (fromTrack && toTrack && !toTrack.locked) {
+        const clipIndex = fromTrack.clips.findIndex(c => c.id === clipId)
+        if (clipIndex !== -1) {
+          const [clip] = fromTrack.clips.splice(clipIndex, 1)
+          clip.start = Math.max(0, newStart)
+          toTrack.clips.push(clip)
+          toTrack.clips.sort((a, b) => a.start - b.start)
+          if (fromTrack.type === 'video') enforceVideoNoOverlap(fromTrack)
+          if (toTrack.type === 'video') enforceVideoNoOverlap(toTrack)
+          state.totalDuration = recalcTotalDuration(state.tracks)
+        }
+      }
+    }),
+
+    groupSelected: (groupId) => set((state) => {
+      const ids = state.selectedClipIds || []
+      if (!ids.length) return
+      const gid = groupId || `grp_${generateId()}`
+      for (const t of state.tracks) {
+        for (const c of t.clips) {
+          if (ids.includes(c.id)) c.groupId = gid
+        }
+      }
+    }),
+
+    ungroupSelected: () => set((state) => {
+      const ids = state.selectedClipIds || []
+      if (!ids.length) return
+      for (const t of state.tracks) {
+        for (const c of t.clips) {
+          if (ids.includes(c.id)) delete c.groupId
+        }
+      }
+    }),
+
+    shiftGroup: (groupId, delta) => set((state) => {
+      if (!groupId || !Number.isFinite(delta) || Math.abs(delta) < 1e-9) return
+      for (const t of state.tracks) {
+        for (const c of t.clips) {
+          if (c.groupId === groupId) {
+            c.start = Math.max(0, (c.start ?? 0) + delta)
+          }
+        }
+        t.clips.sort((a, b) => a.start - b.start)
+      }
+      state.totalDuration = recalcTotalDuration(state.tracks)
+    }),
+
+    splitClip: (trackId, clipId, time) => set((state) => {
+      const track = state.tracks.find(t => t.id === trackId)
+      if (!track || track.locked) return
+
+      const idx = track.clips.findIndex(c => c.id === clipId)
+      if (idx === -1) return
+      const clip = track.clips[idx]
+
+      const splitOffset = time - clip.start
+      const minDur = 0.5
+      if (splitOffset <= minDur || splitOffset >= clip.duration - minDur) return
+
+      const aId = generateId()
+      const bId = generateId()
+      const trimStart = clip.trimStart ?? 0
+
+      const clipA: TimelineClip = {
+        ...clip,
+        id: aId,
+        duration: splitOffset,
+        trimStart,
+        trimEnd: trimStart + splitOffset,
+      }
+
+      const clipBTrimStart = trimStart + splitOffset
+      const clipB: TimelineClip = {
+        ...clip,
+        id: bId,
+        start: clip.start + splitOffset,
+        duration: clip.duration - splitOffset,
+        trimStart: clipBTrimStart,
+        trimEnd: clipBTrimStart + (clip.duration - splitOffset),
+      }
+
+      // replace in place
+      track.clips.splice(idx, 1, clipA, clipB)
+      track.clips.sort((x, y) => x.start - y.start)
+      if (track.type === 'video') enforceVideoNoOverlap(track)
+      state.selectedClipId = clipB.id
+      state.totalDuration = recalcTotalDuration(state.tracks)
+    }),
+
+    // 以红色播放头为准切分：不要求选中片段；会对所有未锁定轨道中“覆盖该时间点”的片段生效
+    splitAtTime: (time) => set((state) => {
+      const t = Number(time)
+      if (!Number.isFinite(t) || t <= 0) return
+      const minDur = 0.5
+
+      for (const track of state.tracks) {
+        if (!track || track.locked) continue
+        const idx = track.clips.findIndex(c => {
+          const s = c.start ?? 0
+          const e = s + (c.duration ?? 0)
+          return s + minDur < t && t < e - minDur
+        })
+        if (idx === -1) continue
+
+        const clip = track.clips[idx]
+        const splitOffset = t - (clip.start ?? 0)
+        if (splitOffset <= minDur || splitOffset >= (clip.duration ?? 0) - minDur) continue
+
+        const aId = generateId()
+        const bId = generateId()
+        const trimStart0 = clip.trimStart ?? 0
+
+        const clipA: TimelineClip = {
+          ...clip,
+          id: aId,
+          duration: splitOffset,
+          trimStart: track.type === 'video' ? trimStart0 : clip.trimStart,
+          trimEnd: track.type === 'video' ? (trimStart0 + splitOffset) : clip.trimEnd,
+        }
+
+        const clipBTrimStart = trimStart0 + splitOffset
+        const clipB: TimelineClip = {
+          ...clip,
+          id: bId,
+          start: (clip.start ?? 0) + splitOffset,
+          duration: (clip.duration ?? 0) - splitOffset,
+          trimStart: track.type === 'video' ? clipBTrimStart : clip.trimStart,
+          trimEnd: track.type === 'video' ? (clipBTrimStart + ((clip.duration ?? 0) - splitOffset)) : clip.trimEnd,
+        }
+
+        if (track.type === 'video') {
+          clampVideoDur(clipA, minDur)
+          clampVideoDur(clipB, minDur)
+        }
+
+        track.clips.splice(idx, 1, clipA, clipB)
+        track.clips.sort((x, y) => (x.start ?? 0) - (y.start ?? 0))
+        if (track.type === 'video') enforceVideoNoOverlap(track)
+        state.selectedClipId = clipB.id
+        state.selectedTrackId = track.id
+      }
+
+      state.totalDuration = recalcTotalDuration(state.tracks)
+    }),
+
+    rippleShiftFrom: (trackId, clipId, delta) => set((state) => {
+      const track = state.tracks.find(t => t.id === trackId)
+      if (!track || track.locked) return
+      const ordered = track.clips.slice().sort((a, b) => a.start - b.start)
+      const idx = ordered.findIndex(c => c.id === clipId)
+      if (idx === -1) return
+      const ids = new Set(ordered.slice(idx).map(c => c.id))
+      for (const c of track.clips) {
+        if (ids.has(c.id)) {
+          c.start = Math.max(0, (c.start ?? 0) + delta)
+        }
+      }
+      track.clips.sort((a, b) => a.start - b.start)
+      if (track.type === 'video') enforceVideoNoOverlap(track)
+      state.totalDuration = recalcTotalDuration(state.tracks)
+    }),
+    
+    addTrack: (track) => set((state) => {
+      state.tracks.push({
+        ...track,
+        id: generateId(),
+      } as Track)
+    }),
+    
+    updateTrack: (trackId, updates) => set((state) => {
+      const track = state.tracks.find(t => t.id === trackId)
+      if (track) {
+        Object.assign(track, updates)
+      }
+    }),
+    
+    deleteTrack: (trackId) => set((state) => {
+      state.tracks = state.tracks.filter(t => t.id !== trackId)
+      state.totalDuration = recalcTotalDuration(state.tracks)
+      if (state.selectedTrackId === trackId) {
+        state.selectedTrackId = null
+      }
+    }),
+    
+    toggleTrackMute: (trackId) => set((state) => {
+      const track = state.tracks.find(t => t.id === trackId)
+      if (track) {
+        track.muted = !track.muted
+      }
+    }),
+    
+    toggleTrackLock: (trackId) => set((state) => {
+      const track = state.tracks.find(t => t.id === trackId)
+      if (track) {
+        track.locked = !track.locked
+      }
+    }),
+
+    toggleTrackCollapse: (trackId) => set((state) => {
+      const track = state.tracks.find(t => t.id === trackId)
+      if (track) {
+        track.collapsed = !track.collapsed
+      }
+    }),
+    
+    pushHistory: (meta) => set((state) => {
+      const newHistory = state.history.slice(0, state.historyIndex + 1)
+      newHistory.push({
+        tracks: JSON.parse(JSON.stringify(state.tracks)),
+        label: meta?.label || '编辑',
+        icon: meta?.icon,
+        ts: Date.now(),
+      })
+      state.history = newHistory.slice(-30) // 最多保留30步
+      state.historyIndex = state.history.length - 1
+    }),
+    
+    undo: () => set((state) => {
+      if (state.historyIndex > 0) {
+        state.historyIndex--
+        state.tracks = JSON.parse(JSON.stringify(state.history[state.historyIndex].tracks))
+        state.totalDuration = recalcTotalDuration(state.tracks)
+      }
+    }),
+    
+    redo: () => set((state) => {
+      if (state.historyIndex < state.history.length - 1) {
+        state.historyIndex++
+        state.tracks = JSON.parse(JSON.stringify(state.history[state.historyIndex].tracks))
+        state.totalDuration = recalcTotalDuration(state.tracks)
+      }
+    }),
+
+    jumpToHistory: (index) => set((state) => {
+      const i = Math.max(0, Math.min(index, state.history.length - 1))
+      if (i < 0 || i >= state.history.length) return
+      state.historyIndex = i
+      state.tracks = JSON.parse(JSON.stringify(state.history[i].tracks))
+      state.totalDuration = recalcTotalDuration(state.tracks)
+    }),
+    
+    reset: () => set(initialState),
+  })
+  })
+)
+

web/src/templates.ts

@@ -0,0 +1,5 @@
+// Back-compat re-export
+export * from './lib/templates'
+
+
+

web/tailwind.config.js

@@ -0,0 +1,55 @@
+/** @type {import('tailwindcss').Config} */
+export default {
+  content: [
+    "./index.html",
+    "./src/**/*.{js,ts,jsx,tsx}",
+  ],
+  theme: {
+    extend: {
+      colors: {
+        // 深色主题 - 专业视频编辑器风格
+        editor: {
+          bg: '#1a1a1a',
+          panel: '#242424',
+          surface: '#2d2d2d',
+          border: '#3d3d3d',
+          hover: '#404040',
+          accent: '#3b82f6',
+          'accent-hover': '#2563eb',
+          text: '#e5e5e5',
+          'text-muted': '#a3a3a3',
+          success: '#22c55e',
+          warning: '#f59e0b',
+          danger: '#ef4444',
+        },
+        track: {
+          video: '#3b82f6',
+          audio: '#22c55e',
+          voiceover: '#8b5cf6',
+          subtitle: '#f59e0b',
+          fancy: '#ec4899',
+          bgm: '#06b6d4',
+        }
+      },
+      fontFamily: {
+        sans: ['Inter', 'system-ui', '-apple-system', 'sans-serif'],
+        mono: ['JetBrains Mono', 'Menlo', 'Monaco', 'monospace'],
+      },
+    },
+  },
+  plugins: [],
+}
+
+
+
+
+
+
+
+
+
+
+
+
+
+

web/tsconfig.json

@@ -0,0 +1,39 @@
+{
+  "compilerOptions": {
+    "target": "ES2020",
+    "useDefineForClassFields": true,
+    "lib": ["ES2020", "DOM", "DOM.Iterable"],
+    "module": "ESNext",
+    "skipLibCheck": true,
+    "moduleResolution": "bundler",
+    "allowImportingTsExtensions": true,
+    "resolveJsonModule": true,
+    "isolatedModules": true,
+    "noEmit": true,
+    "jsx": "react-jsx",
+    "strict": true,
+    "noUnusedLocals": true,
+    "noUnusedParameters": true,
+    "noFallthroughCasesInSwitch": true,
+    "baseUrl": ".",
+    "paths": {
+      "@/*": ["./src/*"]
+    }
+  },
+  "include": ["src"],
+  "references": [{ "path": "./tsconfig.node.json" }]
+}
+
+
+
+
+
+
+
+
+
+
+
+
+
+

web/tsconfig.node.json

@@ -0,0 +1,24 @@
+{
+  "compilerOptions": {
+    "composite": true,
+    "skipLibCheck": true,
+    "module": "ESNext",
+    "moduleResolution": "bundler",
+    "allowSyntheticDefaultImports": true
+  },
+  "include": ["vite.config.ts"]
+}
+
+
+
+
+
+
+
+
+
+
+
+
+
+

web/vite.config.ts

@@ -0,0 +1,39 @@
+import { defineConfig } from 'vite'
+import react from '@vitejs/plugin-react'
+import path from 'path'
+
+export default defineConfig({
+  plugins: [react()],
+  resolve: {
+    alias: {
+      '@': path.resolve(__dirname, './src'),
+    },
+  },
+  server: {
+    port: 3000,
+    proxy: {
+      '/api': {
+        target: 'http://localhost:8000',
+        changeOrigin: true,
+      },
+      '/static': {
+        target: 'http://localhost:8000',
+        changeOrigin: true,
+      },
+    },
+  },
+})
+
+
+
+
+
+
+
+
+
+
+
+
+
+