HY-Motion-1.0

Why we included this project

Teams building 3D character animations, games, or virtual avatars rarely get an open model that turns plain text into skeleton-based motion data, and that is exactly what HY-Motion 1.0 provides. A prompt like "a person squatting and pressing a barbell overhead" yields motion you can drop into standard 3D animation tools rather than a one-off video demo. The project scales Diffusion Transformer text-to-motion generation to the billion-parameter level, backed by a three-stage training recipe of large-scale pretraining, high-quality fine-tuning, and reinforcement learning. Both the full model and a lighter variant come with a CLI for batch inference, a Gradio interface for interactive previews, and Hugging Face weights, so teams can evaluate on modest hardware before scaling up. The documentation also states honest limits: non-humanoid characters and multi-person interactions are not supported.