Tencent HY-Motion 1.0 is an open-source, billion-parameter model that converts text into 3D character animations using the Diffusion Transformer (DiT) architecture and flow matching. This model enhances the capabilities of developers and creators by providing high-fidelity, fluid, and diverse animations that can be easily integrated into existing 3D animation workflows. It features a full-stage training strategy, including pre-training, supervised fine-tuning, and reinforcement learning, to ensure physical plausibility and semantic accuracy across over 200 motion categories. This advancement sets a new standard for instruction-following capability and motion quality in the industry. This matters because it significantly enhances the ability to create complex and realistic 3D animations from natural language, broadening the possibilities for content creation and innovation in digital media.
