tony/video-flow
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

perf(8502): 并行生图(6并发)+超时重试；视频URL直连预览/下载；路径隔离

Browse Source
This commit is contained in:
Tony Zhang
2025-12-17 12:21:22 +08:00
parent ebcf165c3f
commit 1e210ffccf
12 changed files with 1168 additions and 201 deletions
Download Patch File Download Diff File Expand all files Collapse all files

72
modules/composer.py
View File

@@ -11,6 +11,7 @@ from typing import Dict, Any, List, Optional, Union
import config
from modules import ffmpeg_utils, fancy_text, factory, storage
from modules.text_renderer import renderer
from modules import path_utils
logger = logging.getLogger(__name__)
@@ -65,6 +66,7 @@ class VideoComposer:
bgm_path: str = None,
bgm_volume: float = 0.15,
output_name: str = None,
project_id: Optional[str] = None,
upload_to_r2: bool = False
) -> str:
"""
@@ -89,25 +91,27 @@ class VideoComposer:
timestamp = int(time.time())
output_name = output_name or f"composed_{timestamp}"
# Per-project temp dir to avoid cross-project overwrites
temp_root = path_utils.project_compose_dir(project_id, output_name) if project_id else config.TEMP_DIR
logger.info(f"Starting composition: {len(video_paths)} videos")
try:
# Step 1: 拼接视频
merged_path = str(config.TEMP_DIR / f"{output_name}_merged.mp4")
merged_path = str(Path(temp_root) / f"{output_name}_merged.mp4")
ffmpeg_utils.concat_videos(video_paths, merged_path, self.target_size)
self._add_temp(merged_path)
current_video = merged_path
# Step 1.1: 若无音轨，补一条静音底，避免后续滤镜找不到 0:a
silent_path = str(config.TEMP_DIR / f"{output_name}_silent.mp4")
silent_path = str(Path(temp_root) / f"{output_name}_silent.mp4")
ffmpeg_utils.add_silence_audio(current_video, silent_path)
self._add_temp(silent_path)
current_video = silent_path
# Step 2: 添加字幕 (白字黑边，无底框，水平居中)
if subtitles:
subtitled_path = str(config.TEMP_DIR / f"{output_name}_subtitled.mp4")
subtitled_path = str(Path(temp_root) / f"{output_name}_subtitled.mp4")
subtitle_style = {
"font": ffmpeg_utils._get_font_path(),
"fontsize": 60,
@@ -169,7 +173,7 @@ class VideoComposer:
"duration": ft.get("duration", 999)
})
fancy_path = str(config.TEMP_DIR / f"{output_name}_fancy.mp4")
fancy_path = str(Path(temp_root) / f"{output_name}_fancy.mp4")
ffmpeg_utils.overlay_multiple_images(
current_video, overlay_configs, fancy_path
)
@@ -178,13 +182,15 @@ class VideoComposer:
# Step 4: 生成并混合旁白（火山 WS 优先，失败回退 Edge）
if voiceover_text:
vo_out = str(Path(temp_root) / f"{output_name}_vo_full.mp3")
vo_path = factory.generate_voiceover_volcengine(
text=voiceover_text,
voice_type=self.voice_type
voice_type=self.voice_type,
output_path=vo_out,
)
self._add_temp(vo_path)
voiced_path = str(config.TEMP_DIR / f"{output_name}_voiced.mp4")
voiced_path = str(Path(temp_root) / f"{output_name}_voiced.mp4")
ffmpeg_utils.mix_audio(
current_video, vo_path, voiced_path,
audio_volume=1.5,
@@ -195,12 +201,12 @@ class VideoComposer:
elif voiceover_segments:
current_video = self._add_segmented_voiceover(
current_video, voiceover_segments, output_name
current_video, voiceover_segments, output_name, Path(temp_root)
)
# Step 5: 添加BGM（淡入淡出，若 duck 失败会自动退回低音量混合）
if bgm_path:
bgm_output = str(config.TEMP_DIR / f"{output_name}_bgm.mp4")
bgm_output = str(Path(temp_root) / f"{output_name}_bgm.mp4")
ffmpeg_utils.add_bgm(
current_video, bgm_path, bgm_output,
bgm_volume=bgm_volume,
@@ -237,7 +243,8 @@ class VideoComposer:
self,
video_path: str,
segments: List[Dict[str, Any]],
output_name: str
output_name: str,
temp_root: Path,
) -> str:
"""添加分段旁白"""
if not segments:
@@ -254,7 +261,7 @@ class VideoComposer:
audio_path = factory.generate_voiceover_volcengine(
text=text,
voice_type=voice,
output_path=str(config.TEMP_DIR / f"{output_name}_seg_{i}.mp3")
output_path=str(temp_root / f"{output_name}_seg_{i}.mp3")
)
if audio_path:
@@ -270,7 +277,7 @@ class VideoComposer:
# 依次混入音频
current = video_path
for i, af in enumerate(audio_files):
output = str(config.TEMP_DIR / f"{output_name}_seg_mixed_{i}.mp4")
output = str(temp_root / f"{output_name}_seg_mixed_{i}.mp4")
ffmpeg_utils.mix_audio(
current, af["path"], output,
audio_volume=1.0,
@@ -287,7 +294,8 @@ class VideoComposer:
script: Dict[str, Any],
video_map: Dict[int, str],
bgm_path: str = None,
output_name: str = None
output_name: str = None,
project_id: Optional[str] = None,
) -> str:
"""
基于生成脚本和视频映射进行合成
@@ -340,13 +348,30 @@ class VideoComposer:
# 无 background，不加底框
}
# 让花字时长默认跟随镜头（不改 prompt，仅纠正过短/缺失 duration）
start_in_scene = float(ft.get("start_time", 0) or 0.0)
if start_in_scene < 0:
start_in_scene = 0.0
if start_in_scene >= duration:
start_in_scene = 0.0
ft_dur = ft.get("duration", None)
try:
ft_dur_val = float(ft_dur) if ft_dur is not None else None
except Exception:
ft_dur_val = None
# If too short, extend to scene end
if ft_dur_val is None or ft_dur_val < 1.5:
ft_dur_val = max(duration - start_in_scene, 1.5)
# Clamp within scene duration
ft_dur_val = max(0.5, min(ft_dur_val, duration))
fancy_texts.append({
"text": text,
"style": fixed_style,
"x": "(W-w)/2", # 居中
"y": "180", # 上半区域
"start": total_duration + float(ft.get("start_time", 0)),
"duration": float(ft.get("duration", duration))
"start": total_duration + start_in_scene,
"duration": ft_dur_val
})
total_duration += duration
@@ -354,15 +379,16 @@ class VideoComposer:
# 2. 拼接视频
timestamp = int(time.time())
output_name = output_name or f"composed_{timestamp}"
temp_root = path_utils.project_compose_dir(project_id, output_name) if project_id else config.TEMP_DIR
merged_path = str(config.TEMP_DIR / f"{output_name}_merged.mp4")
merged_path = str(Path(temp_root) / f"{output_name}_merged.mp4")
ffmpeg_utils.concat_videos(video_paths, merged_path, self.target_size)
self._add_temp(merged_path)
current_video = merged_path
# 3. 处理整体旁白时间轴 (New Logic)
voiceover_timeline = script.get("voiceover_timeline", [])
mixed_audio_path = str(config.TEMP_DIR / f"{output_name}_mixed_vo.mp3")
mixed_audio_path = str(Path(temp_root) / f"{output_name}_mixed_vo.mp3")
# 初始化静音底轨 (长度为 total_duration)
ffmpeg_utils._run_ffmpeg([
@@ -401,17 +427,17 @@ class VideoComposer:
tts_path = factory.generate_voiceover_volcengine(
text=text,
voice_type=self.voice_type,
output_path=str(config.TEMP_DIR / f"{output_name}_vo_{i}.mp3")
output_path=str(Path(temp_root) / f"{output_name}_vo_{i}.mp3")
)
self._add_temp(tts_path)
# 调整时长
adjusted_path = str(config.TEMP_DIR / f"{output_name}_vo_adj_{i}.mp3")
adjusted_path = str(Path(temp_root) / f"{output_name}_vo_adj_{i}.mp3")
ffmpeg_utils.adjust_audio_duration(tts_path, target_duration, adjusted_path)
self._add_temp(adjusted_path)
# 混合到总音轨
new_mixed = str(config.TEMP_DIR / f"{output_name}_mixed_{i}.mp3")
new_mixed = str(Path(temp_root) / f"{output_name}_mixed_{i}.mp3")
ffmpeg_utils.mix_audio_at_offset(mixed_audio_path, adjusted_path, target_start, new_mixed)
mixed_audio_path = new_mixed # Update current mixed path
self._add_temp(new_mixed)
@@ -425,7 +451,7 @@ class VideoComposer:
})
# 4. 将合成好的旁白混入视频
voiced_path = str(config.TEMP_DIR / f"{output_name}_voiced.mp4")
voiced_path = str(Path(temp_root) / f"{output_name}_voiced.mp4")
ffmpeg_utils.mix_audio(
current_video, mixed_audio_path, voiced_path,
audio_volume=1.5,
@@ -436,7 +462,7 @@ class VideoComposer:
# 5. 添加字幕 (使用新的 ffmpeg_utils.add_multiple_subtitles)
if subtitles:
subtitled_path = str(config.TEMP_DIR / f"{output_name}_subtitled.mp4")
subtitled_path = str(Path(temp_root) / f"{output_name}_subtitled.mp4")
subtitle_style = {
"font": ffmpeg_utils._get_font_path(),
"fontsize": 60,
@@ -455,7 +481,7 @@ class VideoComposer:
# 6. 添加花字
if fancy_texts:
fancy_path = str(config.TEMP_DIR / f"{output_name}_fancy.mp4")
fancy_path = str(Path(temp_root) / f"{output_name}_fancy.mp4")
overlay_configs = []
for ft in fancy_texts:
@@ -477,7 +503,7 @@ class VideoComposer:
# 7. 添加 BGM
if bgm_path:
bgm_output = str(config.TEMP_DIR / f"{output_name}_bgm.mp4")
bgm_output = str(Path(temp_root) / f"{output_name}_bgm.mp4")
ffmpeg_utils.add_bgm(
current_video, bgm_path, bgm_output,
bgm_volume=0.15