This repository provides a single Google Colab notebook that automates setting up the Wan2GP AI video generation platform in a fresh GPU-backed Colab runtime.
Warning: free Colab GPU runtimes usually provide only 15 GB of T4 VRAM, which is too small for most Wan2GP checkpoints; stick to the Wan 2.2 TextImage2Video 5B FastWan model and select 480p to run without memory errors.
Run the notebook top to bottom to clone Wan2GP, install all system and Python dependencies, and launch the Gradio interface from Colab.
- Confirm GPU runtime – prompts you to enable a GPU accelerator before continuing.
- Configure the workspace path and optional persistent data storage – Wan2GP itself runs from fast ephemeral storage in
/content; setUSE_GOOGLE_DRIVE_DATA = Trueonly if you want Google Drive to keep checkpoints, LoRAs, outputs and model caches across Colab restarts. - Download or update Wan2GP – clones the upstream repository into
/content/Wan2GPor pulls the latest changes when it already exists, then links Wan2GP's checkpoint and output folders into the selected data root. - Install system dependencies – installs video and audio libraries required by Wan2GP.
- Install Python dependencies – reuses the PyTorch build already present in the Colab runtime when it is suitable, and installs Wan2GP's requirements.
- Launch Wan2GP – starts the Gradio UI; keep the cell running while you interact with Wan2GP.
- Google account with access to Colab GPU runtimes (free tier works; paid tiers provide more VRAM).
- Stable internet connection while the setup cells run (downloads the Wan2GP repository and model weights).
- Colab may time out inactive sessions; rerun the notebook from the top to rehydrate the environment.
- Keep the final cell running to maintain the public Gradio link; stopping it will terminate the interface.
Issues and pull requests are welcome. If you notice changes in Colab runtimes or Wan2GP dependencies, please open a PR so the notebook stays up to date.
Setup downloads roughly 4 GB less. No models were dropped — Wan2GP's requirements.txt is installed unchanged. Tested on a free-tier Colab T4.
- Reuse the runtime's existing PyTorch instead of force-reinstalling 2.8.0, which re-downloaded every CUDA wheel Colab already had.
- Install requirements with uv rather than pip, for much faster dependency resolution.
- Drop
xformers: Wan2GP never picks it on a T4, and it was the only reason PyTorch was pinned to 2.8.0. - Install an ONNX Runtime build matching Colab's CUDA version, with an automatic CPU fallback so Wan2GP always starts.
- Fix Wan2GP updates being silently skipped after the first run.
- Skip
apt-getwhen the libraries are already present, and cache downloaded packages on Drive when Drive storage is enabled.