# Money Printer Turbo
A topic goes in. A finished short comes out — script, stock footage, voiceover, captions,
music, cut and timed. It is MIT licensed and it runs on your own machine.
Polled live from the GitHub API on 2026-08-28: **118,081 stars · 17,983 forks · MIT ·
created 2024-03-11 · last push 2026-08-28.**
Install it in about ten minutes
git clone https://github.com/harry0703/MoneyPrinterTurbo.git
cd MoneyPrinterTurbo
docker compose -f docker-compose.release.yml up
Open the WebUI it prints. Put one model key in config.toml. That is the whole setup.
Point it at Claude
The repo treats Anthropic as a first-class provider. config.example.toml lines 175-179
carry the block, described in the file itself as *"Anthropic Claude, called through the
OpenAI-compatible endpoint"*:
llm_provider = "anthropic"
anthropic_api_key = "<your Anthropic key>"
anthropic_model_name = "claude-sonnet-5"
It is not tied to any one model — OpenAI, Gemini, DeepSeek, Moonshot and Azure blocks sit in
the same file. Claude is simply one of the supported choices, and if you already pay for it,
it is the only thing this costs you.
The four steps
1. You type a topic. Not a video, not a URL. A subject, in words.
2. It writes the script. The model you configured writes the hook and the body, and
generates the search terms that go with each line.
3. It pulls the footage. Stock clips from Pexels, matched line by line.
4. It builds the short. Voiceover, captions, music bed, mixed and rendered.
What is actually free
Edge TTS is the default voice path and needs no key. Pexels stock footage runs on the free
tier. The repo is MIT. The model key is the only paid input, and it can be one you already
have.
What it will not do
It shoots nothing. Every clip is stock, matched by keyword. Expect to swap some — that is
the honest cost of the speed.
It does not start from someone else's video. The input is a topic you type. It will not
take a competitor's reel, study it frame by frame and write you a version of it. As of
2026-08-28 no free open-source tool does that whole chain; three of the closest were checked
first-hand and each breaks at a different step — zenstory-ai/video-recap-skills narrates
over the source rather than writing something original, Anil-matcha/AI-Youtube-Shorts-Generator
only reranks moments already inside the input video, and this one starts from a topic.
It does not post anything, and it does not reply to anyone. Distribution is your problem.
Rendering is the slow part. On a small or headless machine, expect to deal with
concurrency limits and a headless-Chrome download before you see a finished file.
Get the kit
Comment SHOTS and you get the exact clone-and-run, the Claude config block, and the
honest page on what it will not do.