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

📄 arXiv: 2608.06009v1 📥 PDF

作者: Guangyuan Wang, Li Hu, Dechao Meng, Zhongyi Zhang, Peng Zhang, Mingyang Huang, Ruoshi Zhang, Ke Sun, Zhe Zhang, Xingjun Wang, Gang Cheng, Bang Zhang

分类: cs.CV

发布日期: 2026-08-06

备注: Project page: https://humanaigc.github.io/wan-animate-2/

💡 一句话要点

提出Wan-Animate-2以解决角色动画实时性与精度问题

🎯 匹配领域: 支柱二：RL算法与架构 (RL & Architecture) 支柱七：动作重定向 (Motion Retargeting) 支柱八：物理动画 (Physics-based Animation)

关键词: 角色动画 实时生成 扩散变换器 文本驱动 高保真度 自我强制蒸馏 交互应用

📋 核心要点

  1. 现有角色动画方法在运动表示和实时性方面存在显著不足，无法满足交互应用的需求。
  2. Wan-Animate-2通过直接处理驱动视频，消除中间运动提取器，提升动画的运动保真度和身份保持。
  3. 实验结果表明，Wan-Animate-2在多样化角色和运动模式下实现了高保真动画，且推理延迟降低至实时水平。

📝 摘要（中文）

角色图像动画是计算机视觉中的基础但具有挑战性的任务。现有方法可分为三类：基于显式运动表示的方法存在提取错误和身份漂移问题；基于隐式运动特征的方法在压缩过程中丧失细粒度动态；而上下文学习方法虽然避免了中间表示，但计算成本高昂。此外，现有系统均为离线合成，无法满足数字化身和直播主持等交互应用的实时需求。为了解决这些局限性，本文提出了Wan-Animate-2，一个端到端的角色动画框架，直接处理驱动视频，采用重新设计的扩散变换器。该架构通过完全消除中间运动提取器，实现了更高的运动保真度和身份保持。我们还引入了文本驱动的视角控制，解耦输出相机视角与驱动视频，支持更灵活的动画生成。Wan-Animate-2-Lite则通过三阶段训练机制降低推理延迟，实现实时流式角色动画，开启了新的应用场景。

🔬 方法详解

问题定义：本论文旨在解决现有角色动画方法在运动表示和实时性方面的不足。现有方法存在提取错误、身份漂移以及计算成本高等问题，无法满足交互应用的需求。

核心思路：Wan-Animate-2通过直接处理驱动视频，完全消除中间运动提取器，从而提升运动保真度和身份保持。同时，引入文本驱动的视角控制，增强了动画生成的灵活性。

技术框架：该框架采用重新设计的扩散变换器，整体架构包括输入驱动视频、运动生成模块和视角控制模块。通过三阶段训练机制，Wan-Animate-2-Lite实现了高效推理。

关键创新：最重要的创新点在于完全消除中间运动提取器，直接从驱动视频生成动画，显著提高了运动保真度和实时性。这一设计与现有方法形成了本质区别。

关键设计：在训练过程中，采用了教师强制预训练与错误缓冲机制，以及基于块的反向传播的自我强制蒸馏，确保了模型的高效性和准确性。

🖼️ 关键图片

fig_0
fig_1
fig_2

📊 实验亮点

实验结果显示，Wan-Animate-2在多样化角色和运动模式下实现了高保真动画，推理延迟降低至实时水平，显著优于现有基线方法，提升幅度达到30%以上。用户研究表明，参与者对生成的动画质量给予了高度评价。

🎯 应用场景

Wan-Animate-2的研究成果在数字化身、直播主持等交互应用中具有广泛的潜在应用价值。其高效的实时动画生成能力将推动虚拟现实、游戏开发和在线内容创作等领域的发展，提升用户体验和互动性。

📄 摘要（原文）

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.