Wan-Animate-2 shipped with open base and distillation weights, dropping the motion extractor entirely.

HumanAIGC released inference scripts plus Base and Distillation weights August 7. Driving videos feed directly into a redesigned Diffusion Transformer with no intermediate motion extractor, improving identity preservation, plus text-driven viewpoint control that decouples output camera perspective from the driving video. A Wan-Animate-2-Lite variant cuts inference latency to real-time for streaming character animation. Weights on Hugging Face and ModelScope.