YouTube Shorts from a Prompt, Running Fully Local

What it does

Feed it a topic and MoneyPrinter asks a local Ollama model to write a script and metadata, then assembles the final Short with MoviePy. The current architecture hands off jobs to a background worker backed by Postgres inside Docker, so a restart mid-generation does not scrap the job.

The interesting bit

It keeps the entire AI layer local—no cloud LLM keys or metered API calls. The README is vague about where the raw footage comes from, though an FAQ asking for your TikTok session ID hints that some footage sourcing happens outside the local stack.

Key highlights

• Fully local script and metadata generation via Ollama
• DB-backed job queue (API + worker + Postgres) for crash-safe processing
• MoviePy handles final video assembly
• Cross-platform ImageMagick auto-detection with manual override
• 13.7k stars, though contributions are currently closed

Caveats

• The README explains the AI and queue architecture but never details the full video sourcing workflow; expect to read the docs folder or ask on Discord
• Pull requests are not being accepted for now
• Several local dependencies—Ollama, ImageMagick, and a TikTok session ID—must be in place before the first job runs

Verdict

A decent starting point for developers tinkering with local, automated short-form video factories. Not the right choice if you want an active contributor community or a turnkey SaaS.