The model takes an input video and a character image. In animation mode, the supplied character follows the human motion in the video; in replacement mode, the supplied character replaces the character in that video.
The repository lists the model as a 14B video-to-video release under the Apache 2.0 licence. Wan-AI says it released both the model weights and inference code on September 19, 2025. The model can be downloaded with huggingface-cli; the accompanying Wan2.2 repository specifies PyTorch 2.4 or later and provides an inference workflow. The model card also points to hosted trials on wan.video, ModelScope Studio, and a Hugging Face Space.
The source establishes the availability, stated inputs, modes, licence, and release materials. It does not establish independent quality measurements, hardware requirements for a complete run, inference speed, training-data provenance, or suitability for any particular commercial production workflow.