How to Use MoneyPrinterTurbo (2026 Setup Guide)

MoneyPrinterTurbo is one of the most-starred open-source video tools on GitHub, and most of what you'll find when you search for it is either the raw repo or a YouTube walkthrough. This is the written version: what it actually does, how to get it running, and where it stops.

What MoneyPrinterTurbo actually is

It's a pipeline, not a model. You hand it a subject like "How AI is changing everyday life" and it runs a fixed sequence: an LLM writes the script and extracts keywords, footage is gathered for each part of the script, a text-to-speech engine narrates it, subtitles get timed to the audio, background music is mixed in, and FFmpeg renders the result at 1080x1920 or 1920x1080.

Where the footage comes from is the setting that decides everything else. The README (checked 2026-10-06) lists three sources:

• Free stock from Pexels, Pixabay, or Coverr, matched by keyword. This is the default and what most tutorials show.
• Your own local images and videos, uploaded as material.
• AI-generated clips from paid providers: MiniMax H3 through Metaso, Volcano Engine Ark Seedance, WaveSpeed, OFox, or MuAPI, or images from an OpenAI-compatible image service turned into animated clips. Each clip is a paid job on that provider's account.

Stock keeps the whole thing close to free. The other two cost you either filming time or a per-clip fee. Keep that trade-off in mind through the rest of this guide.

What you need before you start

• Python 3.11. The project asks for 3.11 or later and explicitly recommends 3.11. Newer versions can surprise you with dependency breaks.
• FFmpeg, either auto-downloaded by the project or installed yourself.
• 4 GB RAM minimum, 8 GB recommended, 16 GB if you plan to batch. No GPU required unless you switch subtitles to Whisper, which wants roughly 3 GB of model files.
• An LLM API key. The project supports a long list: OpenAI, Google Gemini, DeepSeek, Kimi/Moonshot, Qwen, Azure OpenAI, VolcEngine Ark, xAI Grok, MiniMax, and MiMo, plus gateways like Cloudflare AI Gateway, ModelScope, AIHubMix, AIML API, EvoLink, Ollama, OneAPI, LiteLLM, Groq, and Pollinations AI. One is enough.
• An AI footage provider account, only if you want generated clips instead of stock or your own uploads.
• Windows users: keep the install path free of spaces, special characters, and non-ASCII characters. This is called out in the docs and it does bite people.

Installing it

The recommended path uses uv:

git clone https://github.com/harry0703/MoneyPrinterTurbo.git
cd MoneyPrinterTurbo
uv python install 3.11
uv sync --frozen

If you'd rather stay on pip:

python3.11 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Docker skips the Python setup entirely:

cd MoneyPrinterTurbo
docker compose -f docker-compose.release.yml up

That gives you a WebUI at http://127.0.0.1:8501 and API docs at http://127.0.0.1:8080/docs. Windows users also have a one-click package on the GitHub Releases page: extract it, run update.bat, then start.bat.

Running it three different ways

WebUI is where most people start. sh webui.sh on macOS and Linux, .\webui.bat on Windows. You get a form: subject, script, voice, subtitle style, aspect ratio, clip duration, music volume.

API service with uv run python main.py, which is what you want if something else is driving it.

CLI, no browser at all:

uv run python cli.py --video-subject "How AI is changing everyday life"

Configuration lives in config.toml, generated from config.example.toml on first run. You can enter API keys through the WebUI's basic settings or edit the file directly.

Voice, subtitles, and the settings that matter

Narration defaults to Edge TTS, which shows up in the WebUI as "Azure TTS V1" and needs no API key at all. If you want something better, the project also supports Azure TTS V2, SiliconFlow, Gemini TTS, MiMo TTS, ElevenLabs, and self-hosted Chatterbox. There's a no-voice mode too.

Subtitles have two providers. The default edge mode reuses TTS timestamps and needs no GPU. The whisper mode runs faster-whisper for better accuracy:

[app]
subtitle_provider = "whisper"

[whisper]
model_size = "large-v3-turbo"

The model downloads from Hugging Face automatically. If that fails, grab it manually and drop it in models/whisper-large-v3/. Fonts live in resource/fonts and you can add your own; subtitle position, color, size, outline, and background are all configurable. Background music sits in resource/songs, and the repo notes that the bundled tracks came from YouTube videos and should be deleted if there's any copyright concern, which is worth acting on before you publish anything commercially.

Using MoneyPrinterTurbo with Claude Code, Cursor, or Codex

The project now ships its own agent skill (version 1.3.2 as of October 2026). If your assistant can read a skill file and run terminal commands, the README's prompt is two lines:

Use this Skill: https://raw.githubusercontent.com/harry0703/MoneyPrinterTurbo/main/docs/skill/SKILL.md
Create a video with the topic "How AI is changing everyday life."

The skill installs MoneyPrinterTurbo with uv under your home folder, asks only for API keys that are missing, generates the video, and returns the path to the MP4. It supports macOS and Windows. Two defaults are worth knowing before you run it:

• Unless you say otherwise, it makes a Chinese-language 9:16 video with Pexels footage and a Chinese Edge TTS voice. Name the language and voice in your prompt if you want English.
• Paid footage always asks first. If you've configured MiniMax H3, Seedance, WaveSpeed, OFox, or MuAPI, the skill stops and asks you to confirm before each paid run, and it never adds the confirmation flag on its own.

If you want finer control than the skill gives you, the older route still works: point your assistant at cli.py, or at the OpenAPI docs served at /docs, and let it call the endpoints directly.

The practical catch: your agent still can't tell you whether the clips look right together. That judgment stays with you, and it's the step that eats the time.

Two errors almost everyone hits

FFmpeg not found. Point at it explicitly:

[app]
ffmpeg_path = "C:\\Users\\path\\ffmpeg.exe"

Too many open files, usually when batching:

ulimit -n 10240

Realistic expectations

MoneyPrinterTurbo is genuinely good at a specific job: high-volume, general-interest faceless shorts where the visuals only need to be thematically relevant. Commentary channels, list videos, motivational content, explainer shorts. It'll batch several variants so you can pick one. That's real leverage and the project deserves its stars, so go star it.

Where it stops depends on the footage setting. Stock footage can illustrate a topic, but it can't show your thing. If you sell a specific hoodie or run a skincare brand, no keyword search returns that clip, because it was never filmed. Uploading your own footage fixes that, but then you're filming first. The AI providers generate new clips from the script's text, which removes the stock mismatch but still won't put your exact product on screen, since they don't take your product photo as input. Stock-based videos also inherit whatever pacing the clips happen to have, which is why a lot of them feel similar even when the scripts differ.

That's not a flaw in the tool. It's the boundary of starting from a topic.

When you need the other kind of video

The opposite approach is to start from a short that already worked and rebuild its structure with your own product in the frame. That's what Riffkit does. You give it one source, a link or an upload or a saved template, and it studies the formula, meaning the hook, the pacing, and the emotional beats, then generates brand-new footage with your product woven in. It never re-uploads the source clip.

It's hosted, so there's no Python, FFmpeg, or provider keys to set up. Output is post-ready: voiceover, captions timed to the audio, a cover frame, and a caption with hashtags, generated natively in any of 9 languages (English, Spanish, Portuguese, Indonesian, German, French, Italian, Japanese, Mandarin). It's billed by the second of finished video, with plans from $15/mo, or $12/mo billed yearly, and every new account gets free credits to try it. A re-roll bills the same way, and you hold full commercial rights to what you make.

The two tools aren't really competing. MoneyPrinterTurbo starts from a topic and costs close to nothing on stock footage; Riffkit starts from a video that already worked and puts your product in it. If you're running a faceless channel that monetizes through affiliate products, you'll probably want both, and the faceless TikTok Shop affiliate playbook covers where each one fits. For a self-hosted tool closer to Riffkit's job, there's also OpenMontage.

If you're weighing this against other options, we compared the whole field by what actually breaks rather than by star count: MoneyPrinterTurbo alternatives, including which popular recommendation has been unmaintained since February 2025 and which one carries an AGPL license nobody mentions.

Get the Riffkit skill

Riffkit runs as an agent skill, so the same assistant you point at MoneyPrinterTurbo can run it too:

Read https://riffkit.ai/SKILL.md and follow the instructions to set up Riffkit.

Then ask in plain language: "riff this TikTok into a video for my product." The skill is open source at github.com/riffkit/skill. Prefer buttons? The same backend runs in your browser at riffkit.ai.