tano2026/omc-fullstack-kit

Short-Form Video Pipeline

YouTube video → viral short clips (TikTok, Reels, Shorts).

Prerequisites

pip install -r "D:/AI Store/digital-marketing-repos/ai-marketing-skills/requirements.txt"
# yt-dlp + ffmpeg required

Usage

Full pipeline (Claude-based segmentation)

python "D:/AI Store/digital-marketing-repos/ai-marketing-skills/short-form-pipeline/scripts/shortform_pipeline.py" \
  --url "https://www.youtube.com/watch?v=VIDEO_ID" \
  --max-clips 3 \
  --output-dir ./output

Standalone clipper (no Claude, heuristic scoring)

python "D:/AI Store/digital-marketing-repos/ai-marketing-skills/short-form-pipeline/scripts/video_clipper.py" \
  --url "https://www.youtube.com/watch?v=VIDEO_ID"

Pipeline Steps

1. Download — yt-dlp fetches video + VTT captions
2. Transcribe — Whisper word-level timestamps
3. Segment — Claude picks 2-5 best 30-60s moments (hook ≥ 7/10)
4. Cut — FFmpeg extracts clips
5. Vertical crop — Layout-aware 16:9 → 9:16 (face detection với MediaPipe)
6. Caption burn — TikTok-style word-highlighted captions

Layout Types (auto-detected)

Layout	Strategy
talking_head	Face-detected center crop
screen_share_overlay	Screen on top, webcam on bottom
side_by_side	Stack screen on top, face on bottom
gallery_view	Active speaker quadrant

Output

• 1080×1920 (9:16 vertical)
• H.264 + AAC
• Word-highlighted captions
• Ready for TikTok/Reels/Shorts

Tips

• FFmpeg error: don't use -c:v copy with -filter_complex
• Caption sync: run Whisper on cut clip, not source
• Long clips: auto-trimmed >90s → 75s
• Requires ANTHROPIC_API_KEY for AI segmentation