[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-72504":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":16,"stars7d":17,"stars30d":18,"stars90d":16,"forks30d":16,"starsTrendScore":16,"compositeScore":19,"rankGlobal":10,"rankLanguage":10,"license":20,"archived":21,"fork":21,"defaultBranch":22,"hasWiki":23,"hasPages":21,"topics":24,"createdAt":10,"pushedAt":10,"updatedAt":25,"readmeContent":26,"aiSummary":27,"trendingCount":16,"starSnapshotCount":16,"syncStatus":28,"lastSyncTime":29,"discoverSource":30},72504,"HY-Motion-1.0","Tencent-Hunyuan\u002FHY-Motion-1.0","Tencent-Hunyuan","HY-Motion model for 3D human motion or 3D character animation generation. ","https:\u002F\u002Fhunyuan.tencent.com\u002Fmotion",null,"Python",2382,196,17,12,0,7,43,66.68,"Other",false,"master",true,[],"2026-06-06 04:05:21","[中文阅读](README_zh_cn.md)\n\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\".\u002Fassets\u002Fbanner.png\" alt=\"Banner\" width=\"100%\">\n\u003C\u002Fp>\n\n\u003Cdiv align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fhunyuan.tencent.com\u002Fmotion\" target=\"_blank\">\n    \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FOfficial%20Site-333399.svg?logo=homepage\" height=\"22px\" alt=\"Official Site\">\n  \u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FTencent-Hunyuan\u002FHY-Motion-1.0\" target=\"_blank\">\n    \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FGitHub-Repo-181717?logo=github&logoColor=white\" height=\"22px\" alt=\"Github Repo\">\n  \u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Fspaces\u002Ftencent\u002FHY-Motion-1.0\" target=\"_blank\">\n    \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F%F0%9F%A4%97%20Demo-276cb4.svg\" height=\"22px\" alt=\"HuggingFace Space\">\n  \u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Ftencent\u002FHY-Motion-1.0\" target=\"_blank\">\n    \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F%F0%9F%A4%97%20Models-d96902.svg\" height=\"22px\" alt=\"HuggingFace Models\">\n  \u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Farxiv.org\u002Fpdf\u002F2512.23464\" target=\"_blank\">\n    \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FReport-b5212f.svg?logo=arxiv\" height=\"22px\" alt=\"ArXiv Report\">\n  \u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fx.com\u002FTencentHunyuan\" target=\"_blank\">\n    \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FHunyuan-black.svg?logo=x\" height=\"22px\" alt=\"X (Twitter)\">\n  \u003C\u002Fa>\n\u003C\u002Fdiv>\n\n\n# HY-Motion 1.0: Scaling Flow Matching Models for 3D Motion Generation\n\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\".\u002Fassets\u002Fteaser.jpg\" alt=\"Teaser\" width=\"100%\">\n\u003C\u002Fp>\n\n\n## 🔥 News\n- **Jan 29, 2026**: 📊 We released the evaluation prompts and code for **SSAE** (Structured Semantic Alignment Evaluation), a VLM-based metric designed to assess the semantic alignment of generated videos. Check the `ssae` directory for usage details!\n- **Dec 30, 2025**: 🤗 We released the inference code and pretrained models of [HY-Motion 1.0](https:\u002F\u002Fhuggingface.co\u002Ftencent\u002FHY-Motion-1.0). Please give it a try via our [HuggingFace Space](https:\u002F\u002Fhuggingface.co\u002Fspaces\u002Ftencent\u002FHY-Motion-1.0) and our [Official Site](https:\u002F\u002Fhunyuan.tencent.com\u002Fmotion)!\n\n\n## **Introduction**\n\n**HY-Motion 1.0** is a series of text-to-3D human motion generation models based on Diffusion Transformer (DiT) and Flow Matching. It allows developers to generate skeleton-based 3D character animations from simple text prompts, which can be directly integrated into various 3D animation pipelines. This model series is the first to scale DiT-based text-to-motion models to the billion-parameter level, achieving significant improvements in instruction-following capabilities and motion quality over existing open-source models.\n\n### Key Features\n- **State-of-the-Art Performance**: Achieves state-of-the-art performance in both instruction-following capability and generated motion quality.\n\n- **Billion-Scale Models**: We are the first to successfully scale DiT-based models to the billion-parameter level for text-to-motion generation. This results in superior instruction understanding and following capabilities, outperforming comparable open-source models.\n\n- **Advanced Three-Stage Training**: Our models are trained using a comprehensive three-stage process:\n\n    - *Large-Scale Pre-training*: Trained on over 3,000 hours of diverse motion data to learn a broad motion prior.\n\n    - *High-Quality Fine-tuning*: Fine-tuned on 400 hours of curated, high-quality 3D motion data to enhance motion detail and smoothness.\n\n    - *Reinforcement Learning*: Utilizes Reinforcement Learning from human feedback and reward models to further refine instruction-following and motion naturalness.\n\n\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\".\u002Fassets\u002Fpipeline.png\" alt=\"System Overview\" width=\"100%\">\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\".\u002Fassets\u002Farch.png\" alt=\"Architecture\" width=\"100%\">\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\".\u002Fassets\u002Fsotacomp.jpg\" alt=\"ComparisonSoTA\" width=\"100%\">\n\u003C\u002Fp>\n\n\n\n\n## 🎁 Model Zoo\n\n**HY-Motion 1.0 Series**\n\n| Model | Description | Date | Size | Huggingface | VRAM (min) |\n|:-------|:-------------|:------:|:------:|:-------------:|:-------------:|\n| **HY-Motion-1.0** | Standard Text2Motion Model | 2025-12-30 | 1.0B | [Download](https:\u002F\u002Fhuggingface.co\u002Ftencent\u002FHY-Motion-1.0\u002Ftree\u002Fmain\u002FHY-Motion-1.0) | 26GB |\n| **HY-Motion-1.0-Lite** | Lightweight Text2Motion Model | 2025-12-30 | 0.46B | [Download](https:\u002F\u002Fhuggingface.co\u002Ftencent\u002FHY-Motion-1.0\u002Ftree\u002Fmain\u002FHY-Motion-1.0-Lite) | 24GB |\n\n*Note*: To reduce GPU VRAM requirements, please use the following settings: `--num_seeds=1`, text prompt with less than 30 words, and motion length less than 5 seconds.  \n*Note*: This table does not includes GPU VRAM requirements for LLM-based prompt engineering feature. If you have sufficient VRAM to run HY-Motion-1.0 model but gradio fails with a VRAM-related error, Run the Gradio application with prompt engineering disabled by setting the environment variable like this: `DISABLE_PROMPT_ENGINEERING=True python3 gradio_app.py`\n\n## 🤗 Get Started with HY-Motion 1.0\n\nHY-Motion 1.0 supports macOS, Windows, and Linux.\n\n\n- [Code Usage (CLI)](#code-usage-cli)\n- [Gradio App](#gradio-app)\n\n\n#### 1. Installation\n\nFirst, install PyTorch via the [official site](https:\u002F\u002Fpytorch.org\u002F). Then install the dependencies:\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002FTencent-Hunyuan\u002FHY-Motion-1.0.git\ncd HY-Motion-1.0\u002F\n# Make sure git-lfs is installed\ngit lfs pull\npip install -r requirements.txt\n```\n\n#### 2. Download Model Weights\nPlease follow the instructions in [ckpts\u002FREADME.md](ckpts\u002FREADME.md) to download the necessary model weights.\n\n### Code Usage (CLI)\n\nWe provide a script for local batch inference, suitable for processing large amounts of prompts.\n\n```bash\n# HY-Motion-1.0\npython3 local_infer.py --model_path ckpts\u002Ftencent\u002FHY-Motion-1.0\n\n# HY-Motion-1.0-Lite\npython3 local_infer.py --model_path ckpts\u002Ftencent\u002FHY-Motion-1.0-Lite\n```\n\n**Common Parameters:**\n- `--input_text_dir`: Directory containing `.txt` or `.json` prompt files.\n- `--output_dir`: Directory to save results (default: `output\u002Flocal_infer`).\n- `--disable_duration_est`: Disable LLM-based duration estimation.\n- `--disable_rewrite`: Disable LLM-based prompt rewriting.\n- `--prompt_engineering_host` \u002F `--prompt_engineering_model_path`: (Optional) Host address \u002F local checkpoint for the Duration Prediction & Prompt Rewrite Module.\n    - **Download**: You can download the Duration Prediction & Prompt Rewrite Module from [Here](https:\u002F\u002Fhuggingface.co\u002FText2MotionPrompter\u002FText2MotionPrompter).\n    - **Note**: If you **do not** set these  parameter, you must also set `--disable_duration_est` and `--disable_rewrite`. Otherwise, the script will raise an error due to host unavailable.\n\n\n### Gradio App\n\nYou can host a [Gradio](https:\u002F\u002Fwww.gradio.app\u002F) web interface on your local machine for interactive visualization:\n\n```bash\npython3 gradio_app.py\n```\nAfter running the command, open your browser and visit `http:\u002F\u002Flocalhost:7860`\n\n\n## Prompting Guide & Best Practices\n\n1. Language & Length: Please use English. For optimal results, keep your prompt under 60 words. For other languages, please use the Text2MotionPrompter to rewrite the prompt. \n\n2. Content Focus: Focus on action descriptions or detailed movements of the limbs and torso.\n\n3. Current Limitations (**NOT** Supported):\n\n - ❌ Non-humanoid Characters: Animations for animals or non-human creatures. \n - ❌ Subjective\u002FVisual Attributes: Descriptions of complex emotions, clothing, or physical appearance. \n - ❌ Environment & Camera: Descriptions of objects, scenes, or camera angles. \n - ❌ Multi-person Interactions: Motions involving two or more people. \n - ❌ Special Modes: Seamless loop or in-place animations. \n\n4. Example Prompts:\n - A person performs a squat, then pushes a barbell overhead using the power from standing up.\n - A person climbs upward, moving up the slope.\n - A person stands up from the chair, then stretches their arms.\n - A person walks unsteadily, then slowly sits down.\n\n\n## 🔗 BibTeX\n\nIf you found this repository helpful, please cite our reports:\n\n```bibtex\n@article{hymotion2025,\n  title={HY-Motion 1.0: Scaling Flow Matching Models for Text-To-Motion Generation},\n  author={Tencent Hunyuan 3D Digital Human Team},\n  journal={arXiv preprint arXiv:2512.23464},\n  year={2025}\n}\n```\n\n## 🤗 Community Integrations\nWe appreciate the community for creating integrations for HY-Motion! Here are some third-party implementations:\n\n- [ComfyUI-HY-Motion1](https:\u002F\u002Fgithub.com\u002Fjtydhr88\u002FComfyUI-HY-Motion1) by [@jtydhr88](https:\u002F\u002Fgithub.com\u002Fjtydhr88)\n\n## Acknowledgements\n\nWe would like to thank the contributors to the [FLUX](https:\u002F\u002Fgithub.com\u002Fblack-forest-labs\u002Fflux), [diffusers](https:\u002F\u002Fgithub.com\u002Fhuggingface\u002Fdiffusers), [HuggingFace](https:\u002F\u002Fhuggingface.co), [SMPL](https:\u002F\u002Fsmpl.is.tue.mpg.de\u002F)\u002F[SMPLH](https:\u002F\u002Fmano.is.tue.mpg.de\u002F), [CLIP](https:\u002F\u002Fgithub.com\u002Fopenai\u002FCLIP), [Qwen3](https:\u002F\u002Fgithub.com\u002FQwenLM\u002FQwen3), [PyTorch3D](https:\u002F\u002Fgithub.com\u002Ffacebookresearch\u002Fpytorch3d), [kornia](https:\u002F\u002Fgithub.com\u002Fkornia\u002Fkornia), [transforms3d](https:\u002F\u002Fgithub.com\u002Fmatthew-brett\u002Ftransforms3d), [FBX-SDK](https:\u002F\u002Fwww.autodesk.com\u002Fdeveloper-network\u002Fplatform-technologies\u002Ffbx-sdk-2020-0), [GVHMR](https:\u002F\u002Fzju3dv.github.io\u002Fgvhmr\u002F), and [HunyuanVideo](https:\u002F\u002Fgithub.com\u002FTencent-Hunyuan\u002FHunyuanVideo) repositories or tools, for their open research and exploration.\n","HY-Motion 1.0 是一个基于Diffusion Transformer (DiT)和流匹配技术的文本到3D人体动作生成模型，能够从简单的文本提示生成基于骨骼的3D角色动画。其核心功能包括通过文本指令生成高质量的3D动画序列，适用于游戏开发、虚拟现实、电影制作等需要动态3D内容的场景。该项目首次将DiT基础模型扩展至十亿参数级别，显著提升了指令跟随能力和动画质量，代表了当前该领域的最先进技术。此外，HY-Motion 1.0还提供了预训练模型和推理代码，方便用户快速集成到现有的3D动画工作流程中。",2,"2026-06-06 03:43:30","high_star"]