Description
Following a YouTube tutorial means scrubbing back and forth: pause, screenshot, rewind, take notes. vid2guide is a Claude Code skill that hands the whole chore to AI: type /vid2guide plus a video URL and it downloads the video, transcribes it, finds the operation steps, grabs screenshots at the right moments and writes a Markdown and PDF tutorial.
A large model reads the subtitles to identify steps with a confidence score, and low-confidence steps get their screenshots checked by a vision model. Everything lands in a folder named after the date and video, with structured step data, the tutorial, the original video and the subtitles. MIT licensed, usable as a CLI too, with a Codex skill version.
One command to a tutorial: Type /vid2guide with a link in Claude Code; download, transcribe, step detection, screenshots and writing are automatic
Whisper transcription: Timestamped subtitles from Whisper, model size selectable, or bring an existing SRT
Steps with confidence: The model finds steps from the subtitles and scores them; low-confidence steps are refined with a vision model
Key-frame screenshots: ffmpeg captures a frame at each step's key timestamp so text and pictures line up
Markdown plus PDF: A Markdown tutorial with screenshots, a PDF with images embedded, and a structured steps.json
Local videos too: Works on local video files as well as YouTube URLs, with optional web search enrichment
Requirements: Needs ffmpeg and the Python dependencies; step analysis uses a Volcengine ARK (Doubao) API key
A large model reads the subtitles to identify steps with a confidence score, and low-confidence steps get their screenshots checked by a vision model. Everything lands in a folder named after the date and video, with structured step data, the tutorial, the original video and the subtitles. MIT licensed, usable as a CLI too, with a Codex skill version.
Features
One command to a tutorial: Type /vid2guide with a link in Claude Code; download, transcribe, step detection, screenshots and writing are automatic
Whisper transcription: Timestamped subtitles from Whisper, model size selectable, or bring an existing SRT
Steps with confidence: The model finds steps from the subtitles and scores them; low-confidence steps are refined with a vision model
Key-frame screenshots: ffmpeg captures a frame at each step's key timestamp so text and pictures line up
Markdown plus PDF: A Markdown tutorial with screenshots, a PDF with images embedded, and a structured steps.json
Local videos too: Works on local video files as well as YouTube URLs, with optional web search enrichment
Requirements: Needs ffmpeg and the Python dependencies; step analysis uses a Volcengine ARK (Doubao) API key
