Installation
This page outlines a simple path to set up an environment suitable for experimenting with Wan 2.2 Animate concepts. Commands and steps are provided as general guidance and may need adaptation for your system.
Prerequisites
- A machine with a recent GPU and drivers installed.
- Python 3.10+ and a virtual environment tool.
- Enough disk space for model weights and sample data.
1. Create a Virtual Environment
python3 -m venv .venv source .venv/bin/activate python -m pip install --upgrade pip
2. Install Dependencies
Install core packages for video processing and model execution. Adjust versions to match your CUDA setup.
pip install torch torchvision torchaudio pip install opencv-python pillow numpy tqdm pyyaml
3. Prepare Assets
- Place a character image in a folder such as
assets/character
. - Place a reference video in
assets/reference
. - Optionally prepare masks for replacement to define generation regions.
4. Configure Inference
Set basic parameters like input paths, output directory, frame size, and duration. Start with short clips and a moderate resolution.
CONFIG= input_image=assets/character/char.png input_video=assets/reference/ref.mp4 output_dir=outputs/run_01 width=720 height=1280 fps=24 replace=false # set to true for replacement mode
5. Run a Test
Run a short test to validate motion following and identity stability.
python scripts/infer.py --image assets/character/char.png --video assets/reference/ref.mp4 --output outputs/run_01 --width 720 --height 1280 --fps 24 --replace false
6. Replacement Mode
For replacement, provide a mask or region hint. The relighting module should adapt tone for better scene fit.
python scripts/infer.py --image assets/character/char.png --video assets/reference/ref.mp4 --output outputs/run_02 --width 720 --height 1280 --fps 24 --replace true --mask assets/reference/subject_mask.png
Tips
- Use clear source images and stable reference clips.
- Test with short segments before scaling up.
- Check lighting and exposure in your reference footage.
Note: This is a generic setup guide intended for education. Adapt steps to your environment.