Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

HY-Motion 1.0: Scaling Flow Matching Models for Text-To-Motion Generation

Yuxin Wen, Qing Shuai, Di Kang, Jing Li, Cheng Wen, Yue Qian, Ningxin Jiao, Changhai Chen, Weijie Chen, Yiran Wang, Jinkun Guo, Dongyue An, Han Liu, Yanyu Tong, Chao Zhang, Qing Guo, Juan Chen, Qiao Zhang, Youyi Zhang, Zihao Yao, Cheng Zhang, Hong Duan, Xiaoping Wu, Qi Chen, Fei Cheng, Liang Dong, Peng He, Hao Zhang, Jiaxin Lin, Chao Zhang, Zhongyi Fan, Yifan Li, Zhichao Hu, Yuhong Liu, Linus, Jie Jiang, Xiaolong Li, Linchao Bao

TL;DR

HY-Motion 1.0 presents a billion-parameter Diffusion Transformer-based framework for text-to-motion generation, achieving state-of-the-art instruction-following and motion quality. It introduces a full-stage training paradigm (large-scale pretraining, high-quality fine-tuning, and reinforcement learning with human feedback and reward models) and a meticulous data pipeline that yields over 200 motion categories. The work demonstrates that data scale drives instruction understanding while data quality drives fidelity, and it provides a robust, open-source baseline for scalable, high-quality 3D motion generation. The combination of diffusion-based flow matching, a dual-stream Transformer architecture, and RL-based alignment offers a practical path toward commercially viable 3D motion generation systems and sets benchmarks for future scalability in the domain.

Abstract

We present HY-Motion 1.0, a series of state-of-the-art, large-scale, motion generation models capable of generating 3D human motions from textual descriptions. 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, delivering instruction-following capabilities that significantly outperform current open-source benchmarks. Uniquely, we introduce a comprehensive, full-stage training paradigm -- including large-scale pretraining on over 3,000 hours of motion data, high-quality fine-tuning on 400 hours of curated data, and reinforcement learning from both human feedback and reward models -- to ensure precise alignment with the text instruction and high motion quality. This framework is supported by our meticulous data processing pipeline, which performs rigorous motion cleaning and captioning. Consequently, our model achieves the most extensive coverage, spanning over 200 motion categories across 6 major classes. We release HY-Motion 1.0 to the open-source community to foster future research and accelerate the transition of 3D human motion generation models towards commercial maturity.

HY-Motion 1.0: Scaling Flow Matching Models for Text-To-Motion Generation

TL;DR

HY-Motion 1.0 presents a billion-parameter Diffusion Transformer-based framework for text-to-motion generation, achieving state-of-the-art instruction-following and motion quality. It introduces a full-stage training paradigm (large-scale pretraining, high-quality fine-tuning, and reinforcement learning with human feedback and reward models) and a meticulous data pipeline that yields over 200 motion categories. The work demonstrates that data scale drives instruction understanding while data quality drives fidelity, and it provides a robust, open-source baseline for scalable, high-quality 3D motion generation. The combination of diffusion-based flow matching, a dual-stream Transformer architecture, and RL-based alignment offers a practical path toward commercially viable 3D motion generation systems and sets benchmarks for future scalability in the domain.

Abstract

We present HY-Motion 1.0, a series of state-of-the-art, large-scale, motion generation models capable of generating 3D human motions from textual descriptions. 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, delivering instruction-following capabilities that significantly outperform current open-source benchmarks. Uniquely, we introduce a comprehensive, full-stage training paradigm -- including large-scale pretraining on over 3,000 hours of motion data, high-quality fine-tuning on 400 hours of curated data, and reinforcement learning from both human feedback and reward models -- to ensure precise alignment with the text instruction and high motion quality. This framework is supported by our meticulous data processing pipeline, which performs rigorous motion cleaning and captioning. Consequently, our model achieves the most extensive coverage, spanning over 200 motion categories across 6 major classes. We release HY-Motion 1.0 to the open-source community to foster future research and accelerate the transition of 3D human motion generation models towards commercial maturity.
Paper Structure (28 sections, 4 equations, 6 figures, 4 tables)

This paper contains 28 sections, 4 equations, 6 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Data
  3. Data Acquisition and Filtering
  4. Data Acquisition.
  5. Motion Processing and Filtering
  6. Motion Captioning
  7. Taxonomy
  8. Model Design
  9. Motion Representation
  10. HY-Motion DiT
  11. Model Architecture.
  12. Model Architecture.
  13. Text Encoders.
  14. Attention Mechanism and Positional Encoding.
  15. Flow Matching Objective.
  16. ...and 13 more sections

Figures (6)

  • Figure 1: Top: Comparison of HY-Motion 1.0 to state-of-the-art text-to-motion models (DART DartControl, LoM chen2024language, GoToZero Fan_2025_ICCV, and MoMask momask). Bottom: Example results generated by HY-Motion 1.0 (retargeted to different characters).
  • Figure 2: Overview of the data processing pipeline.
  • Figure 3: The hierarchy of our motion categories.
  • Figure 4: Overview of the HY-Motion 1.0 framework.
  • Figure 5: Model architecture of our HY-Motion DiT.
  • ...and 1 more figures