Wan-AI Releases Wan-Animate-2, a 14B Character Animation Model
Wan-AI has released Wan-Animate-2, an open-weights 14B character animation framework that generates high-fidelity motion directly from driving videos using a Diffusion Transformer. The release includes base model weights, inference scripts, and a distilled version designed for real-time applications. By eliminating intermediate motion extractors, the model achieves superior identity preservation and motion fidelity. Furthermore, the introduction of a distilled real-time variant (Wan-Animate-2-Lite) lowers the barrier for streaming and interactive character animation. The framework supports text-driven viewpoint control, allowing users to decouple the camera perspective of the output from the original driving video. It is available in multiple versions, including a distilled Diffusers format designed to reduce inference latency to real-time thresholds.
## BACKGROUND
Character animation models often rely on Diffusion Transformers (DiT), which combine diffusion processes with transformer architectures to scale generative performance. To make these large models practical, developers use model distillation to transfer knowledge from a large model to a smaller, faster version, enabling real-time inference.