Wan-Video /Wan-Animate-2

About Wan-Animate-2

Wan-Animate-2 turns a character photo and a driving video into an animated clip. It skips the usual motion extractor, so the model generates motion itself. A text prompt can also change the camera angle away from the driving footage. It wants 8 A800 GPUs, and your prompt should be a Chinese image caption.

Best for: Teams with a multi-GPU server who want to animate a fixed character using motion from another clip.

• Animates a reference character using motion from a driving video
• Skips the motion extractor; motion is generated inside the model
• Changes the output camera angle with a text prompt
• Runs a distilled checkpoint in 10 steps with no extra guidance
• Includes a Gradio demo you can run on your own machine
• Works with Diffusers (WanAnimate2Pipeline) and DiffSynth-Studio

1. 1
   Clone the repo with its submodules: the command below

   git clone --recursive https://github.com/Wan-Video/Wan-Animate-2.git

   Copy
2. 2
   Make a Python 3.11 environment and install PyTorch: the command below, then the next command

   conda create -n wan_animate_2 python==3.11 -y

   Copy

   pip install torch==2.7.0 torchvision==0.22.0 torchaudio==2.7.0 --index-url https://download.pytorch.org/whl/cu126

   Copy
3. 3
   Install the rest: the command below, the next command, pip install -e .

   pip install -r requirements.txt

   Copy

   pip install flash-attn --no-build-isolation

   Copy
4. 4
   Download the weights: the command below

   huggingface-cli download Wan-AI/Wan2.2-Animate-2-14B --local-dir ./ckpts/

   Copy
5. 5
   Write a Chinese caption for your reference image with an LLM, then cd infer and run the command below

   python wan_animate_2_demo.py --prompt "..." --refer-img-file ../examples/demo1/reference.png --refer-video-file ../examples/demo1/template.mp4 --config ./wan_animate_2.yaml

   Copy
6. 6
   Or start the local demo: the command below

   python wan_animate_2_gradio.py

   Copy