SIGGRAPH Asia 2026

UniMate

One Unified Model to Animate Diverse Skeletons
Linzhan Mou, Jiahui Lei, Zhiyang Dou, Chenyue Cai, Chaoyue Song, Adam Finkelstein, Szymon Rusinkiewicz
Princeton University University of California, Berkeley Massachusetts Institute of Technology
arXiv Poster Interactive demo Code Dataset

Given a rigged 3D asset and a text prompt, UniMate generates animations for characters with diverse skeletal topologies within a single unified model.

Interactive Examples.

Drag to orbit Scroll to zoom

Open full screen
Loading…
Drop .fbx / .glb / .gltf to visualize

Abstract.

Recent advances in automatic rigging now deliver animation-ready 3D assets at scale, yet generating the motion to drive them remains a bottleneck. Existing learned animators are topology-constrained: they rely on category-specific templates or require per-skeleton fine-tuning and reference motions at inference. We present UniMate, a unified foundation model that synthesizes articulated motion for arbitrary skeletons from a rigged 3D asset and a text prompt, with no test-time optimization or per-skeleton retraining. UniMate relies on a topology-aware diffusion transformer, which integrates skeletal topology into attention via three mechanisms: (1) a graph-aware attention bias from pairwise joint relations and geodesic distances; (2) a spectral rotary position embedding generalizing RoPE to arbitrary kinematic trees via the graph Laplacian; and (3) a global topological conditioner attention-pooled from the rest-pose skeleton. We also curate UniML3D, 13,006 motion sequences spanning bipedal, quadrupedal, avian, marine, insectoid, serpentine, and articulated rigid objects with unified canonicalization and text pairing. Trained on this dataset, UniMate outperforms state-of-the-art baselines in quality, generalization, and efficiency, and supports zero-shot cross-topology transfer, in-betweening, expansion, and text-guided editing.

Experiments.

1. Diverse Skeletons, Diverse Prompts.
2. One Prompt, Diverse Skeletons.
3. One Skeleton, Diverse Prompts.
4. One Skeleton, One Prompt, Diverse Motions.

Applications.

All three applications are zero-shot: they reuse the same pretrained model with no fine-tuning or auxiliary networks, differing only in which motion tokens are held fixed during sampling.
1. Motion Editing.
2. Motion In-Betweening.
3. Motion Expansion.

Dataset.

Training an animator that generalizes across object categories needs motion data with the same breadth, but raw 4D sources are noisy and inconsistent: disconnected or scene-level skeletons, broken roots, non-functional joints, physically implausible motion, and mismatched coordinate frames or facing directions. We curate UniML3D by unifying three complementary sources — Truebones, Mixamo, and Objaverse-XL.
Sample rigs and motion sequences from UniML3D, spanning bipedal, quadrupedal, avian, marine, insectoid, serpentine, and articulated rigid skeletons.

Samples from UniML3D, spanning bipedal, quadrupedal, avian, marine, insectoid, serpentine, and articulated rigid objects, with detailed skeleton annotations and coherent text prompts paired with motion sequences.

UniML3D is released on Hugging Face, with the data-processing code for the three passes below.
Three passes over every asset. Each line removes (−), adds (+), or re-expresses (→) what the pass was handed.
filter(raw clip) → kept clip
1 prune to a single tree − every tree but the primary one, by cumulative skinning weight
2 realign the root by forward kinematics → a spurious root carried onto the semantic root
3 remove phantom joints − IK controllers and helper bones with zero skinning weight
4 drop static clips − sequences with negligible bone-length-normalized displacement
5 reject implausible clips − outlier root velocity, or jitter past an anatomical threshold
annotate(kept clip) → captioned clip
6 standardize joint names → one anatomical vocabulary, mapped by an LLM
7 select the lateral joint pair + a symmetric pair for facing (thighs, front shoulders), by an LLM
8 resample and render → 30 FPS, four synchronized views, the rest pose rendered alongside
9 caption with a multimodal LLM + one text prompt per clip; “in place” marks a stationary root
10 review every pair by hand − captions that misidentify the dominant action or the moving part
canonicalize(captioned clip) → training sample
11 serialize the tree breadth-first → one joint ordering per skeleton
12 scale by topology diameter → comparable extent across rigs
13 place in a canonical frame → y-axis up, the initial root at the origin
14 align the initial facing direction → +z, from the left-to-right axis of that symmetric pair
15 express rotations against the rest pose → one kinematic basis across topologies
16 normalize by feature statistics → global on the root, local on the rest

Poster.

The UniMate poster: the topology-aware diffusion transformer, generated motions across diverse skeletons, long coherent sequences, in-betweening, editing and expansion, and the UniML3D dataset.

The SIGGRAPH Asia 2026 poster. Click to open it full size, or as a PDF.

BibTeX
@article{mou2026unimate,
  title   = {UniMate: One Unified Model to Animate Diverse Skeletons},
  author  = {Mou, Linzhan and Lei, Jiahui and Dou, Zhiyang and Cai, Chenyue and
             Song, Chaoyue and Finkelstein, Adam and Rusinkiewicz, Szymon},
  journal = {arXiv preprint arXiv:2609.05415},
  year    = {2026}
}