Video
Maestro
A one-click local studio that directs, shoots, and edits AI video for you.
★ 686 starsCustom (non-commercial)
View guide →A lean, fast video generator tuned to squeeze big results out of modest GPUs.
Wan2GP is an open-source video project by deepbeepmeep (deepbeepmeep/Wan2GP on GitHub). It has 9,911 GitHub stars and 1,572 forks, is released under the Custom (WanGP Community 2.0) license, and was last updated on GitHub on 1 October 2026.
Double-click the .bat installer on Windows or run the .sh script on Linux/macOS. It creates the environment and installs the best acceleration kernels for your GPU automatically.
Choose from Wan 2.1/2.2, LTX-2, Hunyuan Video, Qwen Image, or Flux, depending on your VRAM and the look you want.
Enter a text prompt — or an image for image-to-video — and set length, resolution, and steps to match your card.
Optimized attention kernels (Triton, SageAttention and friends) keep render times reasonable even on smaller GPUs.
The whole project is tuned around the “GPU poor”: good results without a 24GB flagship card.
Wan 2.1/2.2, LTX-2, Hunyuan Video, Qwen Image, and Flux behind one consistent interface.
Automated .bat/.sh scripts manage environments, PyTorch, and acceleration kernels for you.
Installs the right Triton, SageAttention, and quantization kernels for your exact GPU and PyTorch combo.
Community Pinokio scripts and a desktop launcher exist if you prefer a guided install.
The same pipeline drives full studio apps like Maestro — start lean here, graduate later.
Download the source with git and move into the folder.
git clone https://github.com/deepbeepmeep/Wan2GP.git
cd Wan2GPMake an isolated Python 3.11 environment with conda.
conda create -n wan2gp python=3.11.14
conda activate wan2gpInstall the CUDA build of PyTorch, then the project requirements. (Or skip all of this and double-click the .bat auto-installer.)
pip install torch==2.10.0 torchvision==0.25.0 torchaudio==2.10.0 --index-url https://download.pytorch.org/whl/cu130
pip install -r requirements.txtWan2GP uses the WanGP Community License 2.0 — a custom license whose plain-English summary says you may use it for free (including inside a company), modify it, and sell the outputs you create. Read LICENSE.txt in the repo for the full terms.
RTX 20-series through 50-series are the documented range, with per-generation install notes. Smaller cards are exactly the target audience.
Maestro builds on this pipeline and wraps it in a full studio with Director mode. Wan2GP is the leaner, faster core for people who want the engine directly.
Yes — the one-click .bat/.sh scripts, community Pinokio scripts, and a desktop launcher by a community member all avoid the manual conda route.
Yes — Qwen Image and Flux support cover image generation alongside the video models.
A one-click local studio that directs, shoots, and edits AI video for you.
A self-hosted creative suite with 600+ image and video models behind one clean interface.
The node-graph powerhouse for precise, professional AI image and video generation.
Free self-hosted AI video generator — multi-scene video with narration.
Feed it a short story and it produces an abstract animated short in the style
A fully automated short-video engine: script, images, voiceover and edit — generated end to end
The most-starred other Video projects in the LibreAI catalogue — the realistic shortlist if Wan2GP is not the right fit.
A feature-rich command-line audio/video downloader
real time face swap and one-click video deepfake with only a single image
OBS Studio - Free and open source software for live streaming and screen recording
Mirror of https://git.ffmpeg.org/ffmpeg.git
World's first open-source, agentic video production system. 12 production pipelines, 100+ tools, 700+ agent skill and production-knowledge files. Turn your A…
Free and open — grab the source, star the repo, and start creating.