Recipe Extraction From Videos — Watch, Transcribe, Save
March 9, 2026 · 2 min read
You find an incredible pasta recipe on YouTube. Twelve-minute video, no written recipe anywhere. Your options? Type it all out, bookmark it and re-watch every time, or just send the link to your agent.
You send a YouTube link. You get back a clean, formatted recipe PDF in your Google Drive. That's the whole interaction.
The Pipeline
Everything runs automatically on your server:
- Download audio via yt-dlp — no cloud service, no subscription
- Transcribe with Whisper (open-source, runs locally) — captures every word
- Extract the recipe — separates ingredients from commentary, organizes steps, notes temps and timing
- Format as PDF — ingredients at top, numbered steps, prep/cook time, servings
- Upload to Google Drive — organized by cuisine, chef, or however you want
A few minutes later: "Done — saved 'Cacio e Pepe from Chef Marco' to your Recipes folder." Click the link. There's your recipe.
Why ChatGPT Can't Do This
ChatGPT can't download videos, run transcription, generate PDFs, or upload to Google Drive. It can't even access YouTube links.
You could cobble together multiple tools manually — one to transcribe, one to extract, one to format. But you'd be the glue, copying and pasting between services.
OpenClaw runs the entire pipeline because it's real software on real hardware. yt-dlp, Whisper, PDF generation, Google Drive API — all installed, all orchestrated automatically.
Works for More Than Cooking
Same pipeline, different content:
- Woodworking tutorials → step-by-step project plans with cut lists
- Lecture recordings → organized study notes
- Conference talks → key points and takeaways
- Craft demonstrations → clean written instructions
Build a Collection Without the Work
Over time, you build a growing library of well-formatted documents in your Google Drive. Searchable. Organized. Ready when you need them.
No more recipe apps with subscriptions. No more screenshot folders. No more bookmarked videos you'll never find again. You send a link, you get a document. Everything in between just works.