Tencent’s Hunyuan team has officially open-sourced ‘HY-Motion 1.0’, a billion-parameter text-to-motion model built on a Diffusion Transformer (DiT) architecture. This innovative model allows animators to generate high-fidelity 3D character skeleton animations from simple natural language descriptions, bypassing the need for manual keyframing or expensive motion capture. The system is designed to directly support mainstream tools such as ‘Blender’, ‘Unity’, and ‘Unreal Engine’, streamlining the production supply chain.

The model was trained on over 3,000 hours of diverse motion data, utilizing a “scale-then-refine” strategy to ensure physical plausibility and smooth transitions. It covers over 200 motion categories, including sports, daily activities, and specific game character actions like sword blocking. This makes ‘HY-Motion 1.0’ a valuable asset for studio houses and game developers looking to populate virtual worlds with realistic NPC behavior and diverse human movements quickly.

Released on December 30, 2025, the model family includes a standard 1B parameter version and a lightweight ‘HY-Motion-1.0-Lite’ for efficient deployment. Technical details reveal the use of the SMPL-H skeleton format and a dual encoder scheme involving ‘Qwen3’ and ‘CLIP-L’ for superior instruction following. By making the code and checkpoints available on GitHub and Hugging Face, Tencent aims to accelerate the industry’s transition toward AI-assisted animation production. (China)