Wan-Animate-2: Pushing the Application Boundaries of Character Animation

2026-08-06

AI Summary

Existing character animation methods struggle to meet real-time interaction demands due to limitations in motion representation, detail preservation, or computational efficiency. This work proposes Wan-Animate-2, an end-to-end framework that eliminates intermediate motion representations and directly processes driving videos within a reconstructed Diffusion Transformer, significantly enhancing identity preservation and motion fidelity. Notably, it achieves the first text-controllable viewpoint disentanglement and supports lightweight streaming inference for real-time generation. Through a three-stage training strategy—comprising teacher-forced pretraining, self-forced distillation, and chunked backpropagation—the method delivers high-fidelity animation across diverse characters and motions. User studies confirm its superiority over existing approaches, and the base model weights have been publicly released.

Abstract

Character image animation remains a foundational yet challenging task in computer vision. Existing approaches can be broadly categorized into three paradigms: methods based on explicit motion representations suffer from extraction errors and identity drift; methods based on implicit motion features lose fine-grained dynamics through compression; and in-context learning approaches avoid intermediate representations but incur prohibitive computational costs. Furthermore, all current systems are designed for offline synthesis, unable to meet the real-time requirements of interactive applications such as digital avatars and live-streaming hosts. To address these limitations, we present Wan-Animate-2, an end-to-end character animation framework that directly consumes the driving video within a redesigned Diffusion Transformer. Our architecture achieves superior motion fidelity and identity preservation by eliminating intermediate motion extractors entirely. We further introduce text driven viewpoint control that decouples the output camera perspective from the driving video--a capability rarely supported by prior character animation methods that rely on explicit motion representations. Beyond generation quality, we present Wan-Animate-2-Lite, an efficient variant that reduces inference latency to real-time thresholds through a three-stage training paradigm: teacher forcing pretraining with error buffer mechanism, and Self-Forcing distillation with chunk-wise backpropagation. This enables streaming character animation for interactive applications, opening new deployment scenarios that were previously infeasible. Qualitative evaluations and user studies demonstrate that Wan-Animate-2 achieves high-fidelity animation results across diverse characters and motion patterns. To foster further research and community development, we will release the Wan-Animate-2-Base model weights to the public.

Problem

• character animation
• real-time interaction
• motion fidelity
• identity preservation
• computational efficiency

Innovation

• Diffusion Transformer
• text-driven viewpoint control
• real-time character animation
• motion extraction-free
• self-forcing distillation