Wan Animate 2 is now available in ComfyUI
The Wan team's new SOTA character animation model. Driving video read directly by the transformer, text-driven viewpoint control, and a real-time Lite variant — lands in ComfyUI with native support.
Wan Animate 2 is now supported natively in ComfyUI. This model uses a novel end-to-end character animation framework that directly consumes driving videos in a redesigned Diffusion Transformer, which achieves high-fidelity motion generation and strong identity preservation by eliminating intermediate motion extractors, with further text-driven viewpoint control to decouple the output camera perspective from the driving video. Wan Animate 2 Lite is the efficient variant of the same model, built for streaming character animation.
Two new nodes ship with them: WanAnimate2ToVideo, the conditioning node that takes your reference character and driving video, and WanAnimate2Cache, an optional node that caches the pose branch to cut generation time roughly in half.
Model Highlights
The driving video goes straight into the transformer. No intermediate motion extractor sits between your source footage and the model. That removes a lossy hand-off — anything the extractor failed to represent, like subtle expression or finger position, was gone before generation started.
Higher-fidelity motion. Reading the driving video directly is what the Wan team points to for the jump in motion quality over the previous release.
Stronger identity preservation. The same architectural change holds the reference character’s appearance more consistently across the clip.
Text-driven viewpoint control. Camera perspective in the output is decoupled from camera perspective in the driving video. The driving clip supplies the performance; the prompt can place the camera somewhere else.
Wan Animate 2 Lite runs at real-time latency. The efficient variant brings inference latency down to real-time thresholds, which puts streaming character animation on the table rather than batch rendering only.
Video extension. Generations can be continued past a single clip, carrying motion across the seam.
Optional pose caching. WanAnimate2Cache reuses the pose branch’s work across sampling steps instead of recomputing it every step, roughly halving generation time in exchange for system memory.
Context window support. Long sequences can be generated in windows, including alongside the cache.
Getting Started
Update ComfyUI to the latest version, or open Comfy Cloud.
Load a template from the Templates panel, or download the workflows below.
Connect your reference character and your driving video, set your output resolution and length, and run.
Optional: use WanAnimate2Cache to cut generation time.
Model weights: 🤗 Comfy-Org/Wan-Animate-2
As always, enjoy creating.