Researchers from Westlake University, Nanjing University, and three other institutions released GaussiAnimate, a framework that automatically rigs and animates 3D Gaussian Splatting assets. The system extracts skeleton structures, binds them to deformable surfaces, and synthesizes new animations by blending existing motions, cutting PSNR error by 17.3% compared to standard rigging baselines.

What Happened

GaussiAnimate introduces a Scaffold-Skin Rigging System that works in three stages. First, it compresses deformable 3D Gaussians into free-form bones that approximate surface deformations. Second, it extracts a kinematic skeleton using Mean Curvature Skeleton analysis. Third, a Partwise Motion Matching (PartMM) algorithm synthesizes new motions by retrieving and blending patterns from existing animation data.

The full pipeline takes roughly 15 minutes for reconstruction on an RTX 4090D, plus just 2 minutes for the rigging stages. It handles clothed humans, quadrupeds, and garments without requiring category-specific training.

Why It Matters

Rigging is one of the most time-consuming steps in 3D production. Traditional workflows require manual skeleton placement, weight painting, and motion retargeting. GaussiAnimate automates the entire chain from reconstructed 3D Gaussians to controllable animated characters, with a two-level structure that captures both rigid skeleton motion and non-rigid surface deformation like clothing and skin.

The results are significant. On novel pose reanimation, GaussiAnimate achieves 17.3% PSNR improvement over Linear Blend Skinning and 21.7% over Bag-of-Bones methods. For motion matching in low-data scenarios, it cuts RMSE by 48.4%. These gains come from the framework's ability to decompose complex deformations into manageable bone-level components.

Key Details

  • Pipeline: Deformable Gaussians to free-form bones to skeleton to controllable animation
  • Speed: 15 minutes reconstruction + 2 minutes rigging on consumer GPU (RTX 4090D)
  • Categories: Works across humans, animals, and garments without retraining
  • Contributors: Westlake University, Nanjing University, University of Hong Kong, Macau University of Science and Technology

What to Do Next

The project page shows visual results across categories. The research paper details the Scaffold-Skin architecture and PartMM algorithm. Code is planned for public release. 3D artists working with Gaussian Splatting workflows should monitor the repository, as this could significantly reduce the manual effort needed to bring reconstructed 3D scenes to life.