Researchers introduced UniMate, a unified model for animating diverse skeletons. It synthesizes motion from a rigged 3D asset and a text prompt. The model uses a topology-aware diffusion transformer. It integrates skeletal topology via graph-aware attention, spectral rotary embeddings, and a global topological conditioner. UniMate was trained on UniML3D, a dataset of 13,006 motion sequences. It supports zero-shot cross-topology transfer, in-betweening, expansion, and text-guided editing. The paper was submitted to SIGGRAPH Asia 2026.
What changed
Existing animators needed per-skeleton fine-tuning and reference motions at inference.
What it unlocks
Generating motion for arbitrary skeletons from a rigged 3D asset and a text prompt without per-skeleton retraining.
- 13,006 motion sequences in UniML3D
- bipedal, quadrupedal, avian, marine, insectoid, serpentine, rigid objects
Sources