"""
MatchMe Studio - Brain Module (Multi-stage Analysis & Script Generation)
"""
import json
import logging
from typing import Dict, Any, List, Optional
from openai import OpenAI

import config

logger = logging.getLogger(__name__)

# Use Volcengine (Doubao) via OpenAI Compatible Interface
client = OpenAI(
    api_key=config.VOLC_API_KEY,
    base_url=config.VOLC_BASE_URL
)

# ============================================================
# Stage 1: Analyze Materials
# ============================================================

ANALYZE_SYSTEM_PROMPT = """你是一位资深短视频创作总监，专精TikTok/抖音爆款内容。

任务：深度分析用户提供的素材和需求，识别产品特性、使用场景、目标人群。

分析维度：
1. 产品/服务核心卖点（从素材中提取视觉特征）
2. 视觉风格特征（颜色、质感、包装）
3. 潜在目标受众
4. 内容调性建议

然后检查是否缺少关键信息，如果缺少，生成2-5个问题帮助完善需求。
每个问题必须与短视频创作直接相关。

输出严格JSON格式：
{
  "analysis": "详细分析结果，包括从素材中识别到的视觉元素...",
  "detected_info": {
    "product": "识别到的产品名称和类型",
    "visual_features": ["视觉特征1", "视觉特征2"],
    "audience": "推测的目标人群",
    "style": "推测的风格"
  },
  "missing_info": ["缺少的信息1", "缺少的信息2"],
  "questions": [
    {
      "id": "q1",
      "text": "问题文字（说明为什么这个问题重要）",
      "options": ["选项A", "选项B", "选项C"],
      "allow_multiple": true,
      "allow_custom": true
    }
  ],
  "ready": false
}

如果信息足够，ready=true，questions为空数组。
"""

def analyze_materials(
    prompt: str,
    image_urls: List[str] = None,
    asr_text: str = ""
) -> Dict[str, Any]:
    """
    Deep analysis of user materials.
    Returns analysis text and questions if info is missing.
    """
    logger.info("Brain: Analyzing materials...")
    
    # Using Vision Model format (Doubao Vision)
    # Input format: messages with content list (text + image_url)
    
    content_parts = [{"type": "text", "text": f"用户需求: {prompt}"}]
    
    if asr_text:
        content_parts.append({"type": "text", "text": f"\n视频原声(ASR转写): {asr_text}"})
    
    if image_urls:
        content_parts.append({"type": "text", "text": "\n用户上传的素材图片（请仔细分析这些图片中的产品特征）:"})
        for url in image_urls:
            content_parts.append({
                "type": "image_url",
                "image_url": {"url": url}
            })

    messages = [
        # Note: Some vision models might not support 'system' role with images well, 
        # but Doubao usually follows standard chat structure.
        # If system prompt fails, prepend it to user content.
        {"role": "system", "content": ANALYZE_SYSTEM_PROMPT},
        {"role": "user", "content": content_parts}
    ]

    try:
        # Use Vision Model for Analysis
        response = client.chat.completions.create(
            model=config.VISION_MODEL_ID, 
            messages=messages,
            temperature=0.7,
            max_tokens=4000
        )
        
        content = response.choices[0].message.content.strip()
        if content.startswith("```"):
            parts = content.split("```")
            if len(parts) > 1:
                content = parts[1]
            if content.startswith("json"): content = content[4:]
            
        return json.loads(content)
        
    except Exception as e:
        logger.error(f"Brain Analyze Error: {e}")
        raise


# ============================================================
# Stage 2: Refine Brief with Answers
# ============================================================

REFINE_SYSTEM_PROMPT = """你是短视频创作总监。
根据原始需求、AI分析结果、用户补充回答，整合为完整的创意简报。

注意：用户选择的风格偏好（如ASMR、剧情、视觉流等）必须作为核心创作方向贯穿整个简报。

输出JSON:
{
  "brief": {
    "product": "产品名称",
    "product_visual_description": "产品视觉描述（颜色、形状、包装、质感等，用于后续图片生成）",
    "selling_points": ["卖点1", "卖点2"],
    "target_audience": "目标人群",
    "platform": "投放平台",
    "style": "视频风格（必须明确，如ASMR/剧情/视觉流等）",
    "style_requirements": "该风格的具体创作要求（如ASMR需要：开盖声、质感特写、无人脸等）",
    "creativity_level": "创意程度",
    "reference": "对标账号/竞品",
    "user_assets_description": "用户上传素材的描述（用于后续继承）"
  },
  "creative_summary": "整体创意概述（50字以内，描述这个视频的核心创意方向）",
  "ready": true
}
"""

def refine_brief(
    original_prompt: str,
    analysis: Dict[str, Any],
    answers: Dict[str, Any],
    image_urls: List[str] = None
) -> Dict[str, Any]:
    """
    Integrate user answers into a complete creative brief.
    """
    logger.info("Brain: Refining brief with answers...")
    
    user_content = f"""
原始需求: {original_prompt}

AI分析结果: {json.dumps(analysis, ensure_ascii=False)}

用户补充回答: {json.dumps(answers, ensure_ascii=False)}

用户上传的素材URL: {json.dumps(image_urls or [], ensure_ascii=False)}
"""

    try:
        # Use Text LLM for reasoning/refining if no new images involved
        # But to keep it simple, we can stick to BRAIN_MODEL_ID (Doubao Pro)
        response = client.chat.completions.create(
            model=config.BRAIN_MODEL_ID,
            messages=[
                {"role": "system", "content": REFINE_SYSTEM_PROMPT},
                {"role": "user", "content": user_content}
            ],
            temperature=0.5,
            max_tokens=3000
        )
        
        content = response.choices[0].message.content.strip()
        if content.startswith("```"):
            parts = content.split("```")
            if len(parts) > 1:
                content = parts[1]
            if content.startswith("json"): content = content[4:]
            
        return json.loads(content)
        
    except Exception as e:
        logger.error(f"Brain Refine Error: {e}")
        raise


# ============================================================
# Stage 3: Generate Script
# ============================================================

SCRIPT_SYSTEM_PROMPT = """你是顶级短视频编导，专精{style}风格内容创作。

根据创意简报生成爆款脚本。必须严格遵循用户选择的风格要求。

脚本结构要求：
1. creative_summary: 整体创意概述（这条视频的核心创意是什么）
2. hook: 前3秒钩子设计（必须抓眼球，符合{style}风格）
3. scenes: 3-8个分镜
4. cta: 结尾行动号召（纯文本字符串）

每个分镜(scene)必须包含：
- id: 分镜编号
- duration: 时长(5/10/15秒，符合视频模型参数)
- timeline: 时间轴 (如 "0:00-0:05")
- image_prompt: 【关键】用于AI生图的详细英文prompt，必须包含：
    * 产品的具体视觉描述（继承自brief中的product_visual_description）
    * 8k, hyper-realistic, cinematic lighting
    * 色调、环境、构图、焦点
    * 风格要求（如ASMR需要：macro shot, satisfying texture, no human face）
- keyframe: {
    "color_tone": "色调",
    "environment": "环境/背景",
    "foreground": "前景元素",
    "focus": "视觉焦点",
    "subject": "主体描述",
    "composition": "构图方式"
  }
- camera_movement: 运镜描述（如：slow zoom in, pan left, static）
- story_beat: 这个分镜在整体故事中的作用
- voiceover: 旁白文字（{style}风格，如ASMR应简短或无旁白，用音效代替）
- sound_design: 音效设计（如：开盖声、水滴声、环境白噪音）
- rhythm: {"change": "保持/加快/放慢", "multiplier": 1.0}

旁白要求：
- 必须连贯，形成完整的叙事
- 符合{style}风格（ASMR风格应极简或无旁白）
- 每句旁白要能独立成句，但连起来是完整故事

输出严格JSON格式。
"""

def generate_script(
    brief: Dict[str, Any],
    image_urls: List[str] = None,
    regenerate_feedback: str = ""
) -> Dict[str, Any]:
    """
    Generate complete video script with scenes.
    """
    logger.info("Brain: Generating script...")
    
    style = brief.get("style", "现代广告")
    system_prompt = SCRIPT_SYSTEM_PROMPT.replace("{style}", style)
    
    content_parts = [{"type": "text", "text": f"创意简报: {json.dumps(brief, ensure_ascii=False)}"}]
    
    if regenerate_feedback:
        content_parts.append({"type": "text", "text": f"\n用户反馈(请据此调整): {regenerate_feedback}"})
    
    if image_urls:
        content_parts.append({"type": "text", "text": "\n用户上传的参考素材（生成的image_prompt必须参考这些素材中的产品外观）:"})
        for url in image_urls:
            content_parts.append({
                "type": "image_url",
                "image_url": {"url": url}
            })

    try:
        response = client.chat.completions.create(
            model=config.VISION_MODEL_ID, # Use Vision model to see reference images if available
            messages=[
                {"role": "system", "content": system_prompt},
                {"role": "user", "content": content_parts}
            ],
            temperature=0.8,
            max_tokens=8000
        )
        
        content = response.choices[0].message.content.strip()
        if content.startswith("```"):
            parts = content.split("```")
            if len(parts) > 1:
                content = parts[1]
            if content.startswith("json"): content = content[4:]
            
        return json.loads(content)
        
    except Exception as e:
        logger.error(f"Brain Script Error: {e}")
        raise


# ============================================================
# Stage 4: Regenerate Single Scene
# ============================================================

def regenerate_scene(
    full_script: Dict[str, Any],
    scene_id: int,
    feedback: str,
    brief: Dict[str, Any] = None
) -> Dict[str, Any]:
    """
    Regenerate a single scene based on feedback.
    """
    logger.info(f"Brain: Regenerating scene {scene_id}...")
    
    style = brief.get("style", "现代广告") if brief else "现代广告"
    
    system_prompt = f"""你是短视频编导，专精{style}风格。根据用户反馈重新生成指定分镜。
保持与其他分镜的风格连贯性。
image_prompt必须继承产品的视觉描述。
只输出新的scene对象(JSON)。
"""

    user_content = f"""
完整脚本: {json.dumps(full_script, ensure_ascii=False)}

创意简报: {json.dumps(brief, ensure_ascii=False) if brief else "无"}

需要重新生成的分镜ID: {scene_id}

用户反馈: {feedback}
"""

    try:
        response = client.chat.completions.create(
            model=config.BRAIN_MODEL_ID,
            messages=[
                {"role": "system", "content": system_prompt},
                {"role": "user", "content": user_content}
            ],
            temperature=0.8,
            max_tokens=2000
        )
        
        content = response.choices[0].message.content.strip()
        if content.startswith("```"):
            parts = content.split("```")
            if len(parts) > 1:
                content = parts[1]
            if content.startswith("json"): content = content[4:]
            
        return json.loads(content)
        
    except Exception as e:
        logger.error(f"Brain Regenerate Scene Error: {e}")
        raise