Video Scraping

Skill Seekers v3.2.0 introduced a complete video tutorial extraction pipeline that converts YouTube videos and local video files into AI-consumable skills.

Quick Start

# YouTube video
skill-seekers video --url https://www.youtube.com/watch?v=... --name mytutorial

# Local video file
skill-seekers video --video-file tutorial.mp4 --name mytutorial

# YouTube playlist
skill-seekers video --playlist https://www.youtube.com/playlist?list=... --name series

# First time setup (installs GPU-aware deps)
skill-seekers video --setup

What Gets Extracted

The video pipeline extracts multiple data streams:

  1. Metadata — Title, channel, views, chapters, duration (via yt-dlp)
  2. Transcripts — Multi-source with 3-tier fallback: YouTube Transcript API → yt-dlp subtitles → faster-whisper local transcription
  3. Segments — Chapter-based or time-window segmentation with configurable overlap
  4. Visual code — OCR on code editor panels, code timeline tracking across frames
  5. SKILL.md — Structured output combining all extracted knowledge

Optional Dependencies

pip install "skill-seekers[video]"

This installs: yt-dlp, faster-whisper, pytesseract, opencv-python, and other visual extraction dependencies.

Pipeline Architecture

The video pipeline follows this flow:

Video → Metadata → Transcript → Segmentation → Visual Extraction → SKILL.md

Key modules:

  • video_scraper.py — Main orchestrator (~960 lines)
  • video_metadata.py — YouTube/local metadata extraction
  • video_transcript.py — Multi-source transcript with fallback
  • video_segmenter.py — Chapter-based and time-window segmentation
  • video_visual.py — Visual extraction with OCR (~2,410 lines)

Also Available via Create

# Auto-detection routes YouTube URLs to video scraper
skill-seekers create https://www.youtube.com/watch?v=...