AI 电商视频生成：产品视频一键制作

用户输入：产品图片 + 产品信息
        ↓
┌──────────────────────────────────────┐
│        AI 电商视频生成引擎             │
│                                      │
│  ┌────────────┐  ┌────────────┐     │
│  │  内容规划    │  │  视觉生成    │     │
│  │  脚本生成    │  │  场景合成    │     │
│  │  分镜设计    │  │  产品植入    │     │
│  └──────┬─────┘  └──────┬─────┘     │
│         └───────┬───────┘           │
│                 ↓                    │
│  ┌──────────────────────────┐       │
│  │      视频生成引擎          │       │
│  │  可灵3.0 / Sora 2 / Runway│       │
│  └──────────┬───────────────┘       │
│             ↓                        │
│  ┌──────────────────────────┐       │
│  │      后期合成引擎          │       │
│  │  音频+字幕+水印+转场       │       │
│  └──────────────────────────┘       │
└──────────────────────────────────────┘
        ↓
输出：成品视频（多平台格式）

PRODUCT_VIDEO_TEMPLATES = {
    "rotate_360": {
        "name": "360度旋转展示",
        "duration": 15,
        "scenes": [
            {"time": "0-5s", "action": "产品正面缓慢旋转", "camera": "固定"},
            {"time": "5-10s", "action": "产品侧面展示", "camera": "环绕"},
            {"time": "10-15s", "action": "产品背面+LOGO", "camera": "固定"},
        ]
    },
    "lifestyle": {
        "name": "使用场景展示",
        "duration": 20,
        "scenes": [
            {"time": "0-3s", "action": "产品特写", "camera": "微距"},
            {"time": "3-8s", "action": "使用场景1", "camera": "中景"},
            {"time": "8-13s", "action": "使用场景2", "camera": "中景"},
            {"time": "13-17s", "action": "产品细节", "camera": "特写"},
            {"time": "17-20s", "action": "品牌LOGO+CTA", "camera": "固定"},
        ]
    },
    "comparison": {
        "name": "对比展示",
        "duration": 15,
        "scenes": [
            {"time": "0-3s", "action": "痛点场景", "camera": "中景"},
            {"time": "3-5s", "action": "产品出现", "camera": "特写"},
            {"time": "5-10s", "action": "使用过程", "camera": "中景"},
            {"time": "10-13s", "action": "效果对比", "camera": "固定"},
            {"time": "13-15s", "action": "购买引导", "camera": "固定"},
        ]
    }
}

class ProductVideoGenerator:
    """电商产品视频生成器"""
    
    def __init__(self):
        self.llm = OpenAI()
        self.video_api = KlingAPI()  # 可灵 3.0
        self.tts = CosyVoice2()
        self.bgm = MusicGen()
    
    async def generate(self, product_image, product_info, template="lifestyle"):
        """
        一键生成产品视频
        product_image: 产品图片路径
        product_info: {"name": "...", "price": "...", "selling_points": [...]}
        template: 模板名称
        """
        # 1. 生成视频脚本
        script = await self._generate_script(product_info, template)
        
        # 2. 生成场景描述
        scene_prompts = self._build_scene_prompts(script, product_info)
        
        # 3. 调用可灵 3.0 生成视频
        video_segments = []
        for i, prompt in enumerate(scene_prompts):
            video = await self.video_api.generate(
                prompt=prompt,
                reference_image=product_image,
                duration=script["scenes"][i]["duration"],
                aspect_ratio="9:16",  # 竖屏适合手机
                style="commercial"
            )
            video_segments.append(video)
        
        # 4. 生成旁白
        narration = await self.tts.synthesize(
            text=script["narration"],
            voice_id="commercial_female",
            emotion="enthusiastic"
        )
        
        # 5. 生成 BGM
        bgm = await self.bgm.generate(
            prompt=f"{product_info['category']} product video, "
                   f"upbeat, modern, 30 seconds",
            duration=30
        )
        
        # 6. 合成最终视频
        final_video = self._compose(
            video_segments=video_segments,
            narration=narration,
            bgm=bgm,
            subtitles=script["subtitles"],
            watermark=product_info.get("brand_logo")
        )
        
        return final_video
    
    async def _generate_script(self, product_info, template):
        """生成视频脚本"""
        template_config = PRODUCT_VIDEO_TEMPLATES[template]
        
        prompt = f"""
        产品信息：
        - 名称：{product_info['name']}
        - 价格：{product_info['price']}
        - 卖点：{', '.join(product_info['selling_points'])}
        - 目标用户：{product_info.get('target_audience', '通用')}
        
        视频模板：{template_config['name']}
        时长：{template_config['duration']}秒
        
        请生成视频脚本，包含：
        1. 每个场景的视觉描述（用于AI视频生成）
        2. 旁白文案（每场景一句话，不超过15字）
        3. 字幕文字
        4. BGM风格建议
        """
        
        response = await self.llm.chat.completions.acreate(
            model="gpt-4o",
            messages=[{"role": "user", "content": prompt}],
            response_format={"type": "json_object"}
        )
        
        return json.loads(response.choices[0].message.content)

class BatchVideoGenerator:
    """批量视频生成"""
    
    async def batch_generate(self, products, template="lifestyle"):
        """批量生成产品视频"""
        tasks = []
        for product in products:
            task = self.generator.generate(
                product_image=product["image"],
                product_info=product["info"],
                template=template
            )
            tasks.append(task)
        
        # 并发生成（可灵 API 支持最多 10 个并发）
        results = await asyncio.gather(*tasks, return_exceptions=True)
        
        return [
            {"product_id": p["id"], "video": r} 
            for p, r in zip(products, results)
            if not isinstance(r, Exception)
        ]

class MultiPlatformExporter:
    """多平台视频导出"""
    
    PLATFORM_SPECS = {
        "taobao": {"resolution": "1080x1080", "fps": 30, "bitrate": "2000k"},
        "douyin": {"resolution": "1080x1920", "fps": 30, "bitrate": "4000k"},
        "xiaohongshu": {"resolution": "1080x1440", "fps": 30, "bitrate": "3500k"},
        "youtube": {"resolution": "1080x1920", "fps": 30, "bitrate": "5000k"},
    }
    
    def export_all(self, source_video, output_dir):
        """导出所有平台版本"""
        results = {}
        for platform, spec in self.PLATFORM_SPECS.items():
            output_path = f"{output_dir}/{platform}.mp4"
            self._convert(source_video, output_path, spec)
            results[platform] = output_path
        return results
    
    def _convert(self, src, dst, spec):
        """使用 FFmpeg 转换"""
        w, h = spec["resolution"].split("x")
        subprocess.run([
            "ffmpeg", "-i", src,
            "-vf", f"scale={w}:{h}:force_original_aspect_ratio=decrease,"
                   f"pad={w}:{h}:(ow-iw)/2:(oh-ih)/2:black",
            "-r", str(spec["fps"]),
            "-b:v", spec["bitrate"],
            "-c:a", "aac", "-b:a", "128k",
            "-movflags", "+faststart",
            dst
        ], check=True)

class ProductInsertion:
    """产品植入优化"""
    
    def insert_product(self, scene_image, product_image):
        """将产品自然植入场景"""
        
        # 1. SAM 2 精确抠图
        product_mask = self.sam2.segment(product_image)
        product_cutout = self.apply_mask(product_image, product_mask)
        
        # 2. 场景分析（找到合适的放置位置）
        scene_analysis = self.gpt4o.analyze(scene_image, 
            "分析这个场景，找到最适合放置产品的位置，返回坐标和大小建议")
        
        # 3. 光照匹配
        product_lit = self.match_lighting(product_cutout, scene_image)
        
        # 4. 阴影生成
        product_with_shadow = self.add_shadow(product_lit, scene_analysis)
        
        # 5. 透视变换
        product_final = self.perspective_transform(
            product_with_shadow, scene_analysis["perspective"]
        )
        
        # 6. 合成
        result = self.composite(product_final, scene_image, scene_analysis["position"])
        
        return result

class VideoABTest:
    """视频 A/B 测试"""
    
    async def generate_variants(self, product, n=5):
        """生成多个视频变体用于测试"""
        variants = []
        
        # 不同模板
        templates = ["rotate_360", "lifestyle", "comparison"]
        
        # 不同风格
        styles = ["professional", "casual", "luxury"]
        
        # 不同音乐
        moods = ["upbeat", "calm", "energetic"]
        
        for i in range(n):
            variant = await self.generator.generate(
                product_image=product["image"],
                product_info=product["info"],
                template=templates[i % len(templates)],
                style=styles[i % len(styles)],
                bgm_mood=moods[i % len(moods)]
            )
            variants.append(variant)
        
        return variants
    
    def select_best(self, variants, metric="click_rate"):
        """根据数据选择最优版本"""
        # 投放 24 小时后根据 CTR 选择
        best = max(variants, key=lambda v: v["metrics"][metric])
        return best

✅ 好的脚本：
"3秒抓住眼球 → 5秒展示痛点 → 7秒产品出现 → 10秒效果展示 → 5秒逼单"

❌ 不好的脚本：
"这是一个产品，它很好用，你可以买来试试"