Tencent open-sources HY-Motion 1.0 for 3D character animation
Tencent just open-sourced HY-Motion 1.0, a 1B+ parameter text-to-motion model built on the Diffusion Transformer architecture and flow matching. It covers over 200 motion categories across 6 major classes and uses a complete pre-training, SFT, and RL loop to generate physically plausible 3D animations ready for typical production pipelines.
HY-Motion 1.0 is Tencent’s open-source text-to-motion model that turns natural language prompts into fluid, diverse 3D character animations. The generated assets plug straight into standard 3D animation pipelines, making it an immediate tool for game developers and solo creators.
How it works
The model is built on a flow-matching Diffusion Transformer (DiT) architecture scaled past 1 billion parameters to raise the ceiling on instruction following and motion quality.
To keep movements physically plausible and true to the prompt, Tencent implemented a full training pipeline spanning pre-training, supervised fine-tuning (SFT), and reinforcement learning (RL). Its motion dataset covers more than 200 categories across 6 major classes. Code is available on GitHub, weights are on Hugging Face, and the technical report is on arXiv.
Why it matters
For indie studios and 3D animators, generating clean character motion from text can eliminate hours of tedious keyframing and reduce reliance on expensive motion capture rigs.
We think bringing reinforcement learning into a 1B-parameter motion DiT is a smart move. Semantic accuracy is relatively easy, but making generated bodies obey physics has always been the real bottleneck. If integration into real-world pipelines is as seamless as claimed, it gives smaller creators a serious production boost.
Sources (1)
- @tencenthunyuan(opens in a new tab)Originalx.com