MoneyPrinterTurbo is an MIT-licensed tool that turns a topic or keyword into a finished short video: script, matched footage, voiceover, subtitles and background music. It sits at about 138,500 GitHub stars and gained 615 in a day, which put it on GitHub's weekly trending list. The repository describes it as an all-in-one AI short video generator, in Chinese, English and Japanese READMEs.
It is popular because it removes the dull work of assembling a video from parts. It also needs honest caveats: it handles your API keys, offers shortcuts that lean on other companies' subscriptions, and can be used to flood platforms with low-effort clips. This post covers how it works, how to run it, and what to check first.
TL;DR: questions people ask first #
| Question | Answer from the English README |
|---|---|
| What does it make? | Short videos with script, footage, voiceover, subtitles, music |
| Formats | 9:16 (1080x1920), 16:9 (1920x1080), 1:1 (1080x1080) |
| License | MIT |
| Stars | About 138,500, with about 22,100 forks |
| Install | python -m pip install mpt , thenmpt webui |
| Other install paths | Windows one-click package, Docker, uv source install |
| Hardware | 4 cores, 4 GB RAM minimum; GPU optional |
| Python | 3.11 or later |
| Free parts | Edge TTS voices, Pexels and Pixabay footage |
| Paid parts | LLM APIs, AI video APIs, some voice services |
What the pipeline does #
You give it a subject. A language model writes a script, or you supply your own. The tool extracts search keywords from the script, fetches matching footage, either stock clips from Pexels, Pixabay and Coverr or AI-generated clips, and lines them up under a voiceover. It then generates subtitles with your choice of font, position and color, adds background music, and renders the video.
The README lists a lot of integrations, which is why the tool feels like a hub. For scripts it names providers including Kimi, OpenAI, Anthropic Claude, Google Gemini, DeepSeek, Alibaba Qwen, Azure OpenAI, xAI Grok and MiniMax, plus gateways and local runtimes such as OpenRouter, Ollama and LiteLLM. For voice it lists Edge TTS (free, no key), Azure Speech, ElevenLabs, Kokoro, Fish Audio and others. For AI video footage it lists services such as MiniMax H3, Seedance, Wan and WaveSpeed. It can also take your own images and clips.
An agent route is documented too: if your coding agent can read a Skill document and run a terminal, you can paste a prompt that points to the repository's skill file and the agent installs, configures and runs the tool. The README says that workflow currently supports macOS and Windows. We cover skills in general in our agent skills guide.
A filmstrip with one frame turning green, representing a generated short video built frame by frame
How to install and run it #
The README's simplest path is the PyPI package, with Python 3.11 or later in a virtual environment.
python3 -m venv .venv
source .venv/bin/activate
python -m pip install mpt
mpt webui
The web interface opens in your browser. For command-line generation, run mpt cli --help. The app creates a config.toml on first launch, where you add API keys in the WebUI settings. By default it stores configuration and generated videos under your user data directory, and you can change that with the MPT_HOME variable.
Other options:
- Windows: download the one-click
.7zpackage from GitHub Releases, runupdate.bat, thenstart.bat. The README warns that the auto-generated source archives lack these files. - Docker:
docker compose -f docker-compose.release.yml uppulls a prebuilt image from GitHub Container Registry. The web interface is on port 8501 and API documentation on port 8080. - Source: clone the repository and use
uvto manage the environment. - Colab: the README links a Google Colab notebook for a no-install trial.
A note on package names: the README states that the official PyPI package is mpt, maintained by the author, and that a package named MoneyPrinterTurbo on PyPI is maintained by a third party. Use the name the README gives.
Hardware and cost #
The README's table lists a 4-core CPU and 4 GB of RAM as the minimum, 6 to 8 cores and 8 GB as recommended, and 8+ cores and 16 GB as optimal. A GPU is optional. It helps with local speech recognition (faster-whisper), video processing and batch generation. If you rely on cloud models, CPU and RAM matter more than the GPU.
Real cost depends on the services you connect, and the README gives no per-video figure. Generating text with a hosted model is usually the cheap step. AI-generated video clips are billed per second by their providers and generally cost more than free stock footage, so check each provider's pricing before a batch run. The README's sponsor sections quote prices for some providers, for example per-second rates for a video model, but those are advertisements by the sponsors and should be checked on each provider's own pricing page.
The sponsor wall #
The README opens with paid sponsor sections. They include model gateways, video API resellers and referral offers, several with "exclusive" credits for users who register through the project's links. This is common for popular open-source tools, and it is not hidden. It does mean the first screen of the README is advertising, and the providers listed are not independent recommendations. Compare prices and terms yourself, and be careful with low-cost reseller APIs that ask for your payment details or upstream keys.
Terms-of-service and data risks #
This is the part the README does not stress, so we do.
1. Your keys are in a config file. The tool stores API keys in config.toml and in its web settings. Treat that file like a password store: do not commit it, restrict the Docker volume, and use separate keys with spending limits.
2. Subscription sign-in shortcuts. The feature list says you can use a ChatGPT subscription through OpenAI Codex with device-code sign-in, and it names a Claude Code subscription among connection options. Using a consumer or coding-agent subscription to power an unrelated automation tool may not be allowed by the provider's terms, and providers can change or enforce limits. Read the plan terms before you rely on it. We covered a similar grey area in free-claude-code and its terms and data risks.
3. Third-party gateways see your prompts. Routing through an aggregator or reseller means your scripts and keywords pass through another company. If your topics are sensitive, use direct provider APIs or a local model.
4. Footage licenses. Stock sites have their own licenses. Check Pexels, Pixabay and Coverr terms for commercial use and attribution. AI-generated clips follow the video provider's terms.
5. Platform rules. The README says finished videos can be published to TikTok, Instagram and YouTube Shorts. Platforms increasingly require AI disclosure and penalize repetitive, low-effort uploads. For example, itch.io removed terabytes of AI-generated content and Nexus Mods now requires AI disclosure. Check the rules of any channel you post to before you automate output.
6. Accuracy. An LLM-written script can state false facts and the matched footage may not show what the voiceover says. Review every video before publishing.
A safe first 30 minutes #
If you want to try it, set it up so a mistake costs little.
- Use a throwaway environment. Run it in a virtual environment or the Docker image, not on a machine that holds production secrets.
- Create limited keys. Make a new API key per provider with a low monthly cap, and revoke it when you finish testing.
- Start with free parts. Use Edge TTS and free stock footage first. Add paid AI video only after you like the script and voiceover.
- Generate one video and read it closely. Check the script for false claims and the footage for mismatches before generating a batch.
- Keep the output local at first. Do not wire up direct publishing until you have checked the platform's AI-disclosure and repetition rules.
- Watch the data folder. Generated files and
config.tomllive in the data directory shown above, so back up or delete them deliberately.
These steps take little time and remove most of the risks in the previous section.
How it compares with other video-agent tools #
MoneyPrinterTurbo is a template pipeline: fixed steps from topic to render. Agent-driven tools take a different path. OpenMontage lets a coding agent plan and produce a video as a project, and HyperFrames renders video from HTML for agents to control. Research systems such as MoViEdit from Adobe and KAIST focus on editing quality. MoneyPrinterTurbo trades flexibility for speed and a point-and-click interface, which suits batch production of simple explainer clips.
Who should use it #
| If you want to | Fit |
|---|---|
| Make quick explainer or faceless clips for your own channel | Good, with human review |
| Learn how a script-to-video pipeline fits together | Good: the code is open and the steps are clear |
| Build a custom branded video product | Possible, but expect to modify code |
| Mass-post unreviewed videos to many accounts | Poor: platform rules and quality risks |
| Handle confidential topics | Only with local models and direct APIs |
What we did not test #
We read the README and repository page but did not install or run the tool, and we did not review its code. Star counts and sponsor offers are as listed on November 4, 2026. We could not verify the quality of the sample videos or any sponsor price.
Related reading #
Details are accurate as of November 4, 2026. Check the repository for current versions and sponsors.