TL;DR: MoneyPrinterTurbo turns a one-line topic into a finished short video — script, stock footage, voiceover, subtitles, background music — and it is genuinely MIT-licensed, at v1.3.8 (October 3, 2026) with 129k GitHub stars. The catch: the default pipeline leans on cloud services (Edge TTS, Pexels footage), so “self-hosted” takes deliberate configuration. Output quality is faceless-channel grade, not editorial grade.

	MoneyPrinterTurbo	ShortGPT	DIY ComfyUI + ffmpeg
Best for	One-click topic-to-Shorts automation	Experimenting with dubbing/translation flows	Full creative control, AI-generated visuals
License	MIT (verified Oct 2026)	MIT	Varies per model — many are non-commercial
Local LLM	Yes — Ollama listed as a supported provider	Not a documented first-class path	N/A (no script stage)
Hardware floor	4-core CPU, 4GB RAM; GPU optional	Similar CPU-bound profile	12–24GB VRAM for current video models
The catch	Stock-footage look; cloud TTS by default	Self-described “experimental,” 8k stars	Hours of node-graph work per template

Honest take: if you want automated faceless Shorts and you accept the stock-footage aesthetic, MoneyPrinterTurbo is the most maintained open-source option by a wide margin — but treat the “fully local” claim as a project you configure, not a default you get.

What does MoneyPrinterTurbo actually do?

MoneyPrinterTurbo (github.com/harry0703/MoneyPrinterTurbo) generates complete short-form videos from a topic or keyword: an LLM writes the script, the app pulls matching stock clips from Pexels or Pixabay (or uses media you upload), a TTS engine narrates it, subtitles are generated and styled, background music is layered in, and ffmpeg/MoviePy renders the result at 1080×1920 (vertical), 1920×1080, or 1080×1080. A Streamlit web UI at 127.0.0.1:8501 drives the whole thing, and batch mode queues up to 100 videos.

The project is one of the biggest open-source AI repos of 2026 by raw popularity: 129.4k stars and 20.3k forks as of October 10, 2026, with only 16 open issues — unusually low for a repo this size, which signals active triage rather than abandonment. Releases are frequent: v1.3.6 (September 2) added native Volcano Engine Seedance support for AI-generated clips instead of stock footage, v1.3.7 (September 13) added word-by-word pop-up subtitles, and v1.3.8 (October 3) added configurable concurrency for footage downloads and clip rendering.

Star counts measure virality, not maturity — a rule this site applies to every repo. The release cadence and low open-issue count are the better health signals here, and both check out as of October 2026.

Is MoneyPrinterTurbo really open source?

Yes — the LICENSE file in the repo is plain MIT (copyright 2024 Harry), verified October 2026. No commercial restrictions, no MAU caps, no “open core” split where the useful parts sit behind a paid tier. The code is Python (3.11+ required), and the full pipeline — script generation, footage matching, TTS orchestration, subtitle alignment, rendering — is in the repo.

The honest qualifier: an MIT app is not the same as an MIT stack. Out of the box, the default configuration calls external services — a cloud LLM for scripts (unless you point it at Ollama), Edge TTS for narration (free, but it is a Microsoft cloud endpoint, not local inference), and Pexels/Pixabay APIs for footage. The app is open source; the default data flow is not private. More on the actually-local configuration below.

How do you install MoneyPrinterTurbo with Docker?

Two commands get you to the web UI:

git clone https://github.com/harry0703/MoneyPrinterTurbo.git
cd MoneyPrinterTurbo
docker compose -f docker-compose.release.yml up
# ... webui  | You can now view your Streamlit app in your browser.
# ... webui  | URL: http://0.0.0.0:8501

Open http://127.0.0.1:8501, add your API keys in the settings panel (LLM provider plus a free Pexels or Pixabay key for stock footage), and generate. A manual install with uv or venv on Python 3.11 works too, and there is a one-click package for Windows.

Hardware requirements are modest because the heavy lifting is video encoding, not inference: the project’s own floor is a 4-core CPU with 4GB RAM, 8 cores and 16GB recommended for comfortable batch work. No GPU is required. A GPU only matters if you run local Whisper transcription for subtitles — faster-whisper with large-v3 (~3GB model) or large-v3-turbo (~1.6GB) — where 4GB+ VRAM helps. A used RTX 3060 12GB ($260–$296 used, verified September 2026) is already overkill for that job; see the faster-whisper vs whisper.cpp vs WhisperX comparison for how the local transcription engines stack up.

Because the stack is CPU-bound, it also runs fine on a small cloud box — a 4-vCPU VPS on Vultr covers single-video generation if you’d rather not keep a machine rendering at home. Skip GPU cloud instances entirely; this workload doesn’t use them.

Can MoneyPrinterTurbo run fully local with Ollama?

Mostly, with caveats per stage. The LLM stage is the easy part: Ollama is a documented provider alongside OpenAI, Claude, Gemini, DeepSeek, Qwen, Groq, OpenRouter, and LiteLLM, so scripts can come from any local model you already serve. The rest of the pipeline splits like this:

Stage	Default	Fully local option
Script	Cloud LLM	Ollama (any local model)
Footage	Pexels/Pixabay API (needs key + network)	Upload your own images/clips
Voiceover	Edge TTS (free, cloud)	Self-hosted Chatterbox or Kokoro TTS; or no-voice mode
Subtitles	—	faster-whisper, model runs locally
Music/fonts	Bundled/local	Local files

So a genuinely offline pipeline exists — Ollama + your own media library + Kokoro + faster-whisper — but it is a configuration you assemble, and you give up the two conveniences that made the tool viral: free stock footage matched to the script, and zero-setup Edge TTS voices. If your reason for self-hosting is privacy (prompts and topics never leave the machine), budget an evening for the TTS swap. If your reason is just “no subscription,” the default setup already delivers that: Edge TTS costs nothing and Pexels/Pixabay keys are free.

One real-world failure mode worth knowing before your first batch run: on first use, faster-whisper downloads its model from Hugging Face, and on flaky connections the download stalls and subtitle generation fails with no obvious error in the UI. The fix is documented — download the model manually and drop it into the models folder. Similarly, batch renders were fully serial before v1.3.8; if long queues crawl, update and raise the new concurrency settings rather than fighting the old version.

How does it compare to ShortGPT and a DIY pipeline?

ShortGPT (RayVentura/ShortGPT, 8k stars, MIT) automates the same category — script, footage, EdgeTTS/ElevenLabs voiceover, captions, MoviePy render — and adds a dubbing/translation engine MoneyPrinterTurbo lacks. But it describes itself as an “experimental AI framework,” carries 75 open issues against a much smaller codebase, and has nothing like MoneyPrinterTurbo’s release cadence. Choose it only if multilingual dubbing is the actual job.

A DIY ComfyUI pipeline is the opposite trade. ComfyUI with a video model generates original visuals instead of stock clips — a categorically better look — but there is no script stage, no TTS orchestration, no subtitle alignment, and current open-weight video models want 12–24GB of VRAM where MoneyPrinterTurbo wants four CPU cores. You are building a studio, not clicking a button.

MoneyPrinterTurbo’s own answer to the stock-footage ceiling is the v1.3.6+ integration with hosted video-generation APIs (Volcano Engine Seedance, MiniMax clips of 4–15 seconds) — which improves visuals but reintroduces a metered cloud dependency, pulling the tool away from the self-hosted story. Pick which compromise you can live with.

When NOT to use MoneyPrinterTurbo

• You expect monetizable quality by default. The name oversells it. Stock-footage montages with synthetic narration are the most saturated content category on YouTube Shorts and TikTok, and platforms increasingly suppress unedited synthetic content. The tool automates production, not distribution or originality.
• You need original visuals. Stock clips matched by keyword look like stock clips matched by keyword. That ceiling is structural unless you pay for the cloud video-gen integrations.
• You want a zero-cloud default. As covered above, local-everything is achievable but is not what docker compose up gives you.
• Long-form content. The pipeline is built around 15–60 second verticals. It is the wrong tool for a 10-minute explainer.

For image-generation work rather than video assembly, InvokeAI or ComfyUI remain the better-fitting FOSS picks, and GPU sizing guides for those workloads live on our sister site runaihome.com.

FAQ

Does MoneyPrinterTurbo need a GPU? No. The project’s stated floor is a 4-core CPU and 4GB RAM; rendering is CPU-bound. A GPU with 4GB+ VRAM only accelerates optional local Whisper subtitle transcription.

Is MoneyPrinterTurbo free for commercial use? Yes — the application is MIT-licensed (verified October 2026). Check the terms of the services you wire into it separately: Pexels and Pixabay have their own content licenses, and cloud TTS/LLM providers have their own usage terms.

Can it generate videos in English? Yes. The project originated in the Chinese-language community (docs lead with Chinese, and several integrated providers are Chinese cloud services), but script generation, TTS voices, and the UI all support English, and an English README is maintained.

Sources

• MoneyPrinterTurbo GitHub repository — license, requirements, provider list (accessed October 10, 2026)
• MoneyPrinterTurbo releases — v1.3.6–v1.3.8 changelogs
• MoneyPrinterTurbo English README — pipeline stages, TTS engines, hardware table
• ShortGPT GitHub repository — alternative comparison
• Pexels API — free stock footage source used by the default pipeline

What self-hosting actually costs

Real cost breakdowns for self-hosted AI: hardware floors, power, maintenance hours, and the honest comparison against paying for it. No spam, unsubscribe anytime.

Email address Subscribe