UniMate shows one model can animate diverse skeletons across topologies
A cross-topology motion model learns from many rigs to generate coherent animations from text prompts without per-skeleton training, but struggles with rare topologies.
- Publication
- arXiv
- Stage
- Preprint
- What we read
- Summary of the paper
- Authors
- Linzhan Mou, Jiahui Lei, Zhiyang Dou, Chenyue Cai, Chaoyue Song, Adam Finkelstein, Szymon Rusinkiewicz
- Universities and research institutions
- Not yet supplied in verified metadata; the Brief does not guess.
Institution metadata: OpenAlex record ↗
What they did and found
Researchers trained UniMate, a topology-aware diffusion model, on UniML3D to synthesize motion for arbitrary skeletons from rigged assets and a text prompt. The model produced prompt-faithful animations across unseen rigs and enabled cross-topology transfer, in-betweening, and editing without per-skeleton training.
Why it matters
A single model can broaden who creates 3D animation, reducing setup time and enabling flexible workflows across humans, animals, and objects.
UniMate uses topology-aware attention and spectral encodings to connect motion with skeletal structure, allowing the model to generalize to unseen rigs by learning shared motion patterns across topologies.
The approach supports cross-topology transfer and editing, but may underperform on very rare morphologies or complex contact scenarios where data are scarce.
Original sources · 1
- UniMate: One Unified Model to Animate Diverse Skeletons ↗arXiv · 2026-09-04
Check the original paper for its authors, methods, version and access terms.
What remains uncertain
Struggles with very rare topologies; foot-ground contact can drift without explicit contact modeling in practice.
What is your take?
Ask a question, add useful context or share a different perspective. Keep the conversation respectful and grounded.