Tencent Unveils HY-Motion 1.0, An Open Source AI Model for Advanced 3D Character Motion
Tencent’s Hunyuan team has once again captured the spotlight in the 3D and animation industry with the official release of HY-Motion 1.0, an open-source AI model designed to generate 3D character motion directly from text prompts. The model is now available for download, including both source code and model weights, via GitHub and Hugging Face. It is distributed under a special permissive license that allows individuals and companies to use, copy, modify, and even deploy it commercially for free, as long as they are not large-scale platforms with more than 100 million monthly active users.
At the core of HY-Motion 1.0 is a Diffusion Transformer or DiT model with 1 billion parameters, combined with a Flow Matching system. This technical foundation enables the AI to generate complex movements that appear smooth, natural, and structurally stable without obvious distortion. The training process was conducted in three major stages. First, the model learned fundamental motion patterns from over 3,000 hours of motion capture data. This was followed by fine tuning using an additional 400 hours of high quality motion data. Finally, Tencent applied RLHF techniques to help the AI better understand human intent and produce motions that more accurately match text instructions.
For users interested in testing the technology, Tencent provides two versions of the model. The standard HY-Motion-1.0 offers the full 1 billion parameter experience, while HY-Motion-1.0-Lite is a lighter variant with 460 million parameters. Running the model locally requires a high-performance laptop or desktop PC with approximately 24 to 26 GB of VRAM, although Tencent has included configuration options that allow users to reduce memory usage depending on their hardware limitations.
In its initial release, HY-Motion 1.0 focuses exclusively on human character motion. It does not yet support animals, creatures, or highly complex emotional expressions, nor does it handle interactions between multiple characters. To improve accuracy, the system integrates large language models such as Gemini 2.5 Pro or Qwen to refine ambiguous prompts into clearer instructions before passing them to the motion generation model.
Alongside HY-Motion 1.0, Tencent also introduced HY World 1.5, also known as WorldPlay. This real time virtual world generation framework supports 720P resolution and can create a wide range of environments, from realistic settings to fantasy worlds. Complementing this is Hunyuan3D Studio 1.1, a comprehensive 3D creation workspace that supports high quality quad based grid generation. Interested users can log in via email and access the tool for free up to 30 times per day.
To conclude, Tencent Cloud has officially launched Tencent Hunyuan 3D Global as part of its push into the international market. General users can generate 3D content for free up to 20 times per day, while enterprise customers who register for API access receive up to 200 free credits to get started. This move marks a significant step toward making high end 3D and animation technology more accessible and cost effective for creators and businesses across the SEA Region and beyond.
Source: cgworld
