HY-Motion 1.0: Scaling Flow Matching for Text-to-Motion
Published:Dec 29, 2025 13:46
•1 min read
•ArXiv
Analysis
This paper introduces HY-Motion 1.0, a significant advancement in text-to-motion generation. It's notable for scaling up Diffusion Transformer-based flow matching models to a billion-parameter scale, achieving state-of-the-art performance. The comprehensive training paradigm, including pretraining, fine-tuning, and reinforcement learning, along with the data processing pipeline, are key contributions. The open-source release promotes further research and commercialization.
Key Takeaways
- •HY-Motion 1.0 is a state-of-the-art text-to-motion generation model.
- •It utilizes a scaled-up Diffusion Transformer-based flow matching approach.
- •The model employs a comprehensive training paradigm including pretraining, fine-tuning, and reinforcement learning.
- •It covers over 200 motion categories across 6 major classes.
- •The model is released open-source to foster research and commercialization.
Reference
“HY-Motion 1.0 represents the first successful attempt to scale up Diffusion Transformer (DiT)-based flow matching models to the billion-parameter scale within the motion generation domain.”