$npx -y skills add wells1137/media-skills --skill video-breakdownA professional video analysis skill powered by a dual-model architecture: ByteDance Seed-2.0-Mini for rapid previews and Google Gemini 2.5 Pro for deep, cinematic-grade analysis. It provides quantitative quality assessments and meticulous shot-by-shot breakdowns (拉片)
| 1 | # Video Breakdown |
| 2 | |
| 3 | A professional video analysis skill powered by a **dual-model architecture**: ByteDance **Seed-2.0-Mini** for rapid previews and Google **Gemini 2.5 Pro** for deep, cinematic-grade analysis. It provides quantitative quality assessments and meticulous shot-by-shot breakdowns (拉片) for content creators, editors, and filmmakers. |
| 4 | |
| 5 | ## Core Capabilities |
| 6 | |
| 7 | | Capability | Description | Use Case | |
| 8 | | :--- | :--- | :--- | |
| 9 | | **Quality Critique** | Scores 7 technical dimensions (resolution, lighting, audio, stability, composition, pacing, overall) on a 1-10 scale with professional commentary. | Evaluate UGC quality; compare video versions; pre-publish QA. | |
| 10 | | **Shot Breakdown (拉片)** | Deconstructs every shot with precise timestamps, shot type, camera movement, subject, action, and narrative function. | Analyze competitor videos; study cinematic techniques; create shot lists. | |
| 11 | | **Content Strategy** | Assesses hook strength, retention curve, platform fit (TikTok/YouTube/Instagram/LinkedIn), and viral potential. | Optimize content for distribution; identify drop-off points; improve engagement. | |
| 12 | |
| 13 | ## Model Selection |
| 14 | |
| 15 | This skill uses two models, selectable via the `model` parameter: |
| 16 | |
| 17 | | Model | ID | Best For | |
| 18 | | :--- | :--- | :--- | |
| 19 | | `quick` | `bytedance-seed/seed-2.0-mini` | Fast previews, cost-sensitive tasks, initial screening | |
| 20 | | `full` (default) | `google/gemini-2.5-pro` | Deep analysis, precise timestamps, cinematic-grade breakdowns | |
| 21 | |
| 22 | ## How It Works |
| 23 | |
| 24 | The skill calls a hosted proxy service that routes requests to OpenRouter, which dispatches to the selected model. The response is synchronous — the full analysis result is returned directly in the API response. |
| 25 | |
| 26 | ### Workflow |
| 27 | |
| 28 | 1. **Agent**: Calls `POST /api/analyze` with `video_url`, `analysis_type`, and optionally `model`. |
| 29 | 2. **Proxy**: Forwards the request to OpenRouter with the selected model. |
| 30 | 3. **Model**: Analyzes the video and returns structured JSON. |
| 31 | 4. **Agent**: Presents the parsed result to the user. |
| 32 | |
| 33 | ## Usage |
| 34 | |
| 35 | ### 1. Quick Quality Assessment (Seed-2.0-Mini) |
| 36 | |
| 37 | **Goal**: Get a fast quality report for a video. |
| 38 | |
| 39 | **Agent Action**: |
| 40 | |
| 41 | ```json |
| 42 | { |
| 43 | "tool": "video-breakdown.analyze", |
| 44 | "args": { |
| 45 | "video_url": "https://example.com/my-video.mp4", |
| 46 | "analysis_type": "quality_critique", |
| 47 | "model": "quick" |
| 48 | } |
| 49 | } |
| 50 | ``` |
| 51 | |
| 52 | ### 2. Deep Shot-by-Shot Analysis (Gemini 2.5 Pro) |
| 53 | |
| 54 | **Goal**: Get a professional, frame-accurate shot breakdown. |
| 55 | |
| 56 | **Agent Action**: |
| 57 | |
| 58 | ```json |
| 59 | { |
| 60 | "tool": "video-breakdown.analyze", |
| 61 | "args": { |
| 62 | "video_url": "https://example.com/scene.mp4", |
| 63 | "analysis_type": "shot_breakdown", |
| 64 | "model": "full" |
| 65 | } |
| 66 | } |
| 67 | ``` |
| 68 | |
| 69 | **Expected Output**: |
| 70 | |
| 71 | ```json |
| 72 | [ |
| 73 | { |
| 74 | "shot_number": 1, |
| 75 | "start_time": "00:00", |
| 76 | "end_time": "00:04", |
| 77 | "duration_seconds": 4, |
| 78 | "shot_type": "Medium Shot", |
| 79 | "camera_movement": "Static", |
| 80 | "subject": "Young woman walking toward camera", |
| 81 | "action": "Subject walks confidently, making direct eye contact", |
| 82 | "narrative_function": "Establishes protagonist and sets confident tone", |
| 83 | "audio_notes": "Upbeat music begins, no dialogue" |
| 84 | } |
| 85 | ] |
| 86 | ``` |
| 87 | |
| 88 | ### 3. Content Strategy Analysis |
| 89 | |
| 90 | **Goal**: Evaluate a video's social media performance potential. |
| 91 | |
| 92 | **Agent Action**: |
| 93 | |
| 94 | ```json |
| 95 | { |
| 96 | "tool": "video-breakdown.analyze", |
| 97 | "args": { |
| 98 | "video_url": "https://example.com/reel.mp4", |
| 99 | "analysis_type": "content_strategy", |
| 100 | "model": "full" |
| 101 | } |
| 102 | } |
| 103 | ``` |
| 104 | |
| 105 | ## Backend Service API Reference |
| 106 | |
| 107 | The proxy service is deployed on Vercel Pro (300s timeout). |
| 108 | |
| 109 | ### `POST /api/analyze` |
| 110 | |
| 111 | Submits a video for analysis. |
| 112 | |
| 113 | **Request Body**: |
| 114 | |
| 115 | ```json |
| 116 | { |
| 117 | "video_url": "string (required)", |
| 118 | "analysis_type": "quality_critique | shot_breakdown | content_strategy (required)", |
| 119 | "model": "quick | full (optional, default: full)" |
| 120 | } |
| 121 | ``` |
| 122 | |
| 123 | **Response**: |
| 124 | |
| 125 | ```json |
| 126 | { |
| 127 | "model_used": "google/gemini-2.5-pro", |
| 128 | "analysis_type": "shot_breakdown", |
| 129 | "result": { ... } |
| 130 | } |
| 131 | ``` |
| 132 | |
| 133 | ### `GET /api/health` |
| 134 | |
| 135 | Returns service status and available models. |
| 136 | |
| 137 | ## Deployment |
| 138 | |
| 139 | The proxy service requires one environment variable: |
| 140 | |
| 141 | ``` |
| 142 | OPENROUTER_API_KEY=<your-openrouter-api-key> |
| 143 | ``` |
| 144 | |
| 145 | Deploy to Vercel from the `proxy/` directory within this skill. |