MoZoo: Unleashing Video Diffusion Power in Animal Fur and Muscle Simulation

Researchers introduced MoZoo, a diffusion-based solver that turns coarse animal meshes into realistic video with fur and muscle motion.

The system uses role-aware positional encoding and asymmetric attention to improve temporal and structural consistency across varied animal shapes.

They also released MoZoo-Data and MoZooBench, a synthetic-to-real dataset pipeline and a 120-pair benchmark for evaluation.

The work could reduce manual simulation effort and help overcome training-data bottlenecks in AI-driven animal animation.