TubeSleuth

Short-video pipeline on the OpenAI Assistants API: a scriptwriter, an architect that maps the voiced script to timed shots, and a photographer that writes image prompts, around text to speech, Whisper timestamps, image generation, a Remotion render with word-by-word captions and mood-matched music, and YouTube API upload.

Run as a growth experiment from December 2023 to January 2024: 205 Shorts in 34 days at under $1 each, 145K views and 1,000 subscribers from the Shorts feed, then stopped.

How it runs

npm run create opens a menu to create scripts, create videos or upload videos for a batch of ideas from a Notion database.

1. Scripts: the scriptwriter assistant (instructions in src/config/) writes each script.
2. Voiceover: src/agents/1-voiceover.js
3. Timestamps: src/agents/2-scribe.js
4. Images: the architect and photographer assistants plan the shots and prompts; src/agents/3-painter.js generates them.
5. Render: src/agents/4-stitcher.js and src/remotion/
6. Upload: src/agents/5-uploader.js (YouTube Data API)

Notes

• Archived. Not maintained.
• No credentials are included. Bring your own API keys and Google OAuth client; .gitignore keeps them out of the repo.
• The commit history between December 15, 2023 and the final version is not included.