blendshapes

AnyTalk generates 3D speech animations for arbitrary characters without requiring animation data by adapting video diffusion models through character-specific fine-tuning and optimizing blendshape parameters, with a real-time distilled variant.