Overview
UniMate is a groundbreaking research project from SIGGRAPH Asia 2026, presenting a unified model capable of animating a diverse range of skeletons with unprecedented generality. This repository provides the code, dataset, and tools to replicate and extend UniMate's capabilities, paving the way for truly universal text-to-animation.
Key Features
- Unify Diverse Skeletons: UniMate animates bipedal, quadrupedal, avian, marine, insectoid, serpentine, and articulated rigid objects within a single, cohesive framework.
- Large-Scale Dataset (UniML3D): Introduces UniML3D, a massive dataset of 13,006 text-paired motion sequences, all canonicalized for seamless model training across varied topologies.
- Flexible Training Configurations: Offers extensive options for configuring training, including different data combinations, attention mechanisms (
graph_adaln,full_cross_attn), and model depths, allowing for fine-tuned performance. - Efficient Inference & Applications: Enables real-time animation generation for arbitrary skeletons from text prompts, with no per-skeleton retraining. Supports advanced applications like motion in-betweening, text-guided motion editing, and motion expansion.
- Interactive Demo & Pre-trained Models: Features an interactive demo for exploring animation results and provides pre-trained checkpoints on Hugging Face for immediate use.
Typical Use Cases
- 3D Animation Pipelines: Automate and accelerate the generation of character animations for games, films, and virtual reality by simply providing text descriptions.
- Robotics and Simulation: Generate realistic motion for diverse robotic platforms and simulated agents, adaptable to custom or unfamiliar skeletal structures.
- Creative AI Tools: Empower artists and designers with tools to create complex animations without deep technical expertise in motion capture or rigging.
- Research and Development: Serve as a foundational model for further research in generative animation, cross-domain character animation, and controllable motion synthesis.
