中文文档 | English

🎬 One-stop video content extraction and copywriting analysis tool. Download videos, smart subtitle extraction (embedded/burned/audio), and analyze scripts using three AI frameworks.

# 1. yt-dlp (video downloader)
pip install yt-dlp

# 2. FFmpeg (must be installed and in PATH)
ffmpeg -version

# 3. Python dependencies
pip install pysrt python-dotenv

# 4. FunASR (Recommended for Chinese, lightweight & accurate)
pip install funasr modelscope

# 5. RapidOCR (ONNX lightweight, for burned subtitle detection)
pip install rapidocr-onnxruntime

# 6. Whisper (Alternative for English/multilingual)
pip install openai-whisper

# 7. requests (for Douyin download)
pip install requests

This is a Claude Skill designed for AI agents. Install it in your .agent/skills/ directory:

git clone https://github.com/ALBEDO-TABAI/video-copy-analyzer.git .agent/skills/video-copy-analyzer

Then use it with Claude:

"Analyze this video: https://www.bilibili.com/video/BV1xxxxx"

The skill automatically selects the best extraction method:

Video Input
    ↓
[1️⃣ Embedded Subtitle] ──→ Detected ──→ Direct Extract (Highest Accuracy)
    ↓ Not detected
[2️⃣ Burned Subtitle OCR] ──→ RapidOCR Frame Sampling ──→ Detected ──→ Full Video OCR
    ↓ Not detected
[3️⃣ Audio Transcription] ──→ FunASR (Chinese optimized) / Whisper (Multilingual)
    ↓
Output SRT Subtitles

• RapidOCR (ONNX): Lightweight OCR for burned subtitle detection

  • 🚀 Lightweight: ONNX Runtime, no GPU required
  • 🎯 Cross-platform: Windows/Linux/Mac
  • 📦 Easy deploy: Single pip install
  • ✨ High accuracy: Based on PaddleOCR

• FunASR Nano: Alibaba open-source Chinese ASR model

  • 🚀 Lightweight: ~100MB vs Whisper Large ~1.5GB
  • 🎯 Chinese optimized: Better than Whisper for Chinese
  • ⏱️ Timestamp: Word-level timestamps
  • 💨 Fast: Runs well on CPU

• Narrative structure breakdown
• Rhetorical device identification
• Keyword extraction

• Viral-5D Diagnosis: Hook / Emotion / Peaks / CTA / Social Currency
• Style positioning
• Optimization suggestions

• Core value decomposition
• 2-3 creative direction exploration
• Incremental verification points

video-copy-analyzer/
├── SKILL.md                          # Core skill instructions
├── scripts/
│   ├── download_douyin.py            # Douyin video downloader (watermark-free)
│   ├── extract_subtitle_funasr.py    # Smart subtitle extraction (FunASR + RapidOCR)
│   ├── extract_subtitle.py           # Whisper-based extraction
│   ├── transcribe_audio.py           # Audio transcription script
│   └── check_environment.py          # Environment verification
└── references/
    └── analysis-frameworks.md        # Analysis framework details

On first use, the skill will prompt you to set a default output directory:

• Option A: Use default ~/video-analysis/
• Option B: Specify each time
• Option C: Set a fixed custom directory

After analysis, you'll receive:

This is a Claude Skill that works with AI coding assistants:

💡 Best Performance: Tested with Claude 4.5 Opus, achieving optimal results in transcription correction and three-dimensional analysis.

MIT License