# MoneyPrinterTurbo MoneyPrinterTurbo (MPT) is an open-source tool that turns a topic or a keyword into a finished short video. You type "How AI is changing everyday life", and it writes the script, finds matching footage, generates a voiceover, adds subtitles and background music, then renders an HD MP4 in 9:16, 16:9 or 1:1. It's written in [[Python]], MIT licensed, and one of the most starred AI video repos on GitHub. The name tells you who it was built for: people who want to fill TikTok, Reels and YouTube Shorts channels without filming anything. Others had the idea first: FujiwaraChoki's "MoneyPrinter" (same idea, YouTube Shorts from a topic) hit the Hacker News front page in February 2024, a month before MPT's repo was created. The MPT README doesn't mention a link between the two. ## The problem it solves A faceless short takes a surprising number of fiddly steps: write a script, hunt for B-roll, record or synthesize narration, time the captions, pick music, cut it all together and export. Each step is easy. Doing all of them 30 times a month is not. MPT chains them into one pipeline you drive from a web UI, an API, a CLI or an AI agent. ## How it works It's an orchestrator. It does NOT generate every frame with a video model (unless you plug one in). The default flow: 1. **Script**: an LLM writes the narration from your topic (or you paste your own script) and proposes search keywords for the footage 2. **Footage**: those keywords query free stock libraries (Pexels, Pixabay, Coverr), or MPT uses your own local images and clips 3. **Voice**: a [[Text-to-Speech (TTS)|TTS]] provider reads the script. The default is Edge TTS (free, no API key) 4. **Subtitles**: timed from the TTS timestamps (fast, no GPU), or transcribed locally with `faster-whisper` ([[Whisper]] `large-v3`, about 3 GB) when you need a more accurate timeline 5. **Assembly**: footage, voice, subtitles and music are composited and encoded with [[FFmpeg]] 6. **Publishing** (optional): finished videos can be pushed to TikTok, Instagram and YouTube Shorts through a third-party upload service (Upload-Post) ## Key features - **Four ways to drive it**: Streamlit web UI (port 8501), REST API with OpenAPI docs (port 8080), a pure CLI with JSON/JSONL batch manifests (up to 100 tasks), and a published agent skill (`docs/skill/SKILL.md`) that lets an AI agent install it and make a video for you - **Many LLM providers**: OpenAI, Anthropic, Gemini, DeepSeek, Qwen, Kimi, Grok, MiniMax, OpenRouter, LiteLLM, [[Ollama]] for local models, even a Claude Code subscription - **Many voices**: Edge TTS, Azure, Gemini, MiniMax, [[ElevenLabs]], Fish Audio, self-hosted Chatterbox and Kokoro, and VoxCPM with reference-audio [[Voice Cloning|voice cloning]] (added in v1.3.8) - **Optional AI-generated footage**: Seedance, MiniMax H3, Wan and other text-to-video models through paid gateways, when stock clips don't cut it - **Styled subtitles** (font, color, outline, position) and background music with its own volume control - **Batch variants** and task history - **Local video projects** (v1.3.8, CLI only): replace a scene, the narration or the music and re-render only what changed ## Install and usage - **Windows**: a one-click `.7z` package from the GitHub Releases page (`update.bat`, then `start.bat`) - **macOS / Linux**: `git clone`, then `uv sync --frozen` and `sh webui.sh` (Python 3.11+) - **Docker**: `docker compose -f docker-compose.release.yml up` (prebuilt image on GHCR), with a GPU compose file too - **Google Colab** notebook if you just want to try it - CLI one-liner: `uv run python cli.py --video-subject "How AI is changing everyday life"` No GPU required. Minimum is 4 CPU cores and 4 GB RAM; a GPU helps with local Whisper and batch work. The software is free. What you pay for is the LLM (unless you run a local model), any premium voice or AI-footage provider, and your time reviewing the output. ## License and maturity - **License**: MIT (code only; see caveats below for the media) - **Created**: 2024-03-11. Show HN on 2024-11-15 got 2 points and no comments, so most of the traction came from elsewhere (it's especially big in the Chinese-speaking community; the main README is in Chinese) ## Who's behind it GitHub user **harry0703** ("Harry", account since 2013). A second maintainer, rudycelekli. Harry also maintains MangoDisk, an open-source disk cleaner for macOS and Windows. The project is funded by sponsors, and the README shows it: the top third is a wall of sponsored blocks and referral links for Chinese API gateways and model resellers (Kimi, BytePlus/VolcEngine, APIMart, Metaso and others). Keep that in mind when the docs nudge you toward a provider. A hosted version at moneyprinterturbo.app has stopped accepting new generations (it points to a separate service) and isn't part of the repo. ## Limits and caveats - **Footage quality is the weak point.** Stock search runs on keywords, so you get clips that match a word but not the sentence: generic offices, skylines, close-ups of plants. The HN thread on the original MoneyPrinter said exactly that ("blurry closeup plants and half a parrot"; one user found only 1 in 5 clips related to the prompt). Wavect's 2026 review makes the same point about MPT - **The script is only as good as the LLM, and nobody fact-checks it.** A mass-produced explainer can confidently state invented facts - **YouTube monetization risk.** In July 2025 YouTube renamed its "repetitious content" policy to "inauthentic content" to make clear that mass-produced, repetitive videos aren't eligible for the Partner Program. A template where only the keyword changes is the textbook case - **Rights are your problem.** MIT covers the code, not the media. Pexels and Pixabay have their own licenses (no implied endorsement, care with recognizable people and brands). The bundled background tracks in `resource/songs` are, per the README, "music from YouTube videos" that you should delete "if there are copyright issues". On a monetized channel, delete them anyway and bring licensed music - **Voice cloning**: the release notes themselves warn to only use your own voice or recordings you're authorized to use - **Security**: the web UI and API are developer tools with no authentication. Don't expose them to the internet - **EU AI Act** (Article 50) transparency duties can apply to synthetic media, depending on what you publish ## Relevance for my content (YouTube, courses) I teach. My videos work because I'm in them, explaining things I actually use, with my own screen recordings. A stock-footage short with a synthetic voice is the opposite of that, and it would hurt the trust that sells courses. So I won't use MPT to make my channel. Where it could earn a place: - **Repurposing**: feed it an article or newsletter section as a custom script, point it at my own local footage instead of stock, and get a rough vertical teaser in minutes. Then fix it in a real editor - **Topic testing**: cheap drafts to see whether a hook works before I invest a recording session - **Learning material**: the code is a readable map of an AI video pipeline (script, search terms, TTS, Whisper, FFmpeg compositing, API + CLI + agent surfaces). Worth reading if you're building your own video tooling Always with a human review gate before anything gets published. ## Alternatives - **MoneyPrinter** and **MoneyPrinterV2** (FujiwaraChoki): the original open-source take on the same idea - **[[Remotion]]**: code-driven video (React). More work, full control, good for branded explainers - **[[HeyGen]]**: hosted AI avatar videos when you want a presenter instead of stock footage - **[[Sora]]** / **[[Veo 3]]**: generative video models for original footage (MPT can call some of these through gateways) - **[[CapCut]]** / **[[Descript]]**: editors with AI features, for when you record yourself and want speed rather than full automation ## References - GitHub repository: https://github.com/harry0703/MoneyPrinterTurbo - English README (features, install, TTS, subtitles, background music note): https://github.com/harry0703/MoneyPrinterTurbo/blob/main/README-en.md - Releases (v1.3.8 notes, 2026-10-03; release history): https://github.com/harry0703/MoneyPrinterTurbo/releases - License (MIT, "Copyright (c) 2024 Harry"): https://github.com/harry0703/MoneyPrinterTurbo/blob/main/LICENSE - Agent skill: https://raw.githubusercontent.com/harry0703/MoneyPrinterTurbo/main/docs/skill/SKILL.md - GitHub API (stars, forks, created date, contributors, author profile), queried 2026-10-03: https://api.github.com/repos/harry0703/MoneyPrinterTurbo - Show HN: Generate short videos with one click using AI LLM (2024-11-15): https://news.ycombinator.com/item?id=42150493 - Hacker News discussion of the original MoneyPrinter (2024-02-07, 117 points): https://news.ycombinator.com/item?id=39295526 - Original MoneyPrinter: https://github.com/FujiwaraChoki/MoneyPrinter - MoneyPrinterV2: https://github.com/FujiwaraChoki/MoneyPrinterV2 - Wavect, "MoneyPrinterTurbo Review 2026" (2026-08-20; independent review by a dev agency that sells pipeline integration): https://wavect.io/blog/moneyprinterturbo-review-2026/ - YouTube channel monetization policies ("inauthentic content" update, 2025-07-15): https://support.google.com/youtube/answer/1311392 - moneyprinterturbo.app (hosted service, generation retired): https://moneyprinterturbo.app/ ## Related - [[Text-to-Speech (TTS)]] - [[Whisper]] - [[FFmpeg]] - [[Ollama]] - [[Remotion]] - [[HeyGen]] - [[YouTube (MoC)]] - [[Creator Economy]]