Hunyuan3D Studio：面向游戏级 3D 资产生成的端到端 AI 管线
计算机视觉与模式识别
2025-09-17 v1
摘要
高质量 3D 资产的创建是现代游戏开发的基石，长期以来以劳动密集型和专业化的工作流程为特征。本文介绍了 Hunyuan3D Studio，这是一个端到端的 AI 驱动内容创作平台，旨在通过自动化和简化游戏级 3D 资产的生成来革新游戏制作管线。Hunyuan3D Studio 的核心是将一套先进的神经模块（如部件级 3D 生成、多边形生成、语义 UV 等）集成到一个统一且用户友好的系统中。这个统一框架允许将单张概念图像或文本描述快速转化为一个完全实现、生产质量的 3D 模型，并具备优化的几何结构和高度逼真的 PBR 纹理。我们证明，由 Hunyuan3D Studio 生成的资产不仅视觉上引人注目，而且符合当代游戏引擎的严格技术要求，显著减少了迭代时间并降低了 3D 内容创作的门槛。通过提供从创意意图到技术资产的无缝桥梁，Hunyuan3D Studio 代表了游戏开发和交互式媒体中 AI 辅助工作流程的重大飞跃。
引用
@article{arxiv.2509.12815,
title = {Hunyuan3D Studio: End-to-End AI Pipeline for Game-Ready 3D Asset Generation},
author = {Biwen Lei and Yang Li and Xinhai Liu and Shuhui Yang and Lixin Xu and Jingwei Huang and Ruining Tang and Haohan Weng and Jian Liu and Jing Xu and Zhen Zhou and Yiling Zhu and Jiankai Xing and Jiachen Xu and Changfeng Ma and Xinhao Yan and Yunhan Yang and Chunshi Wang and Duoteng Xu and Xueqi Ma and Yuguang Chen and Jing Li and Mingxin Yang and Sheng Zhang and Yifei Feng and Xin Huang and Di Luo and Zebin He and Puhua Jiang and Changrong Hu and Zihan Qin and Shiwei Miao and Haolin Liu and Yunfei Zhao and Zeqiang Lai and Qingxiang Lin and Zibo Zhao and Kunhong Li and Xianghui Yang and Huiwen Shi and Xin Yang and Yuxuan Wang and Zebin Yao and Yihang Lian and Sicong Liu and Xintong Han and Wangchen Qin and Caisheng Ouyang and Jianyin Liu and Tianwen Yuan and Shuai Jiang and Hong Duan and Yanqi Niu and Wencong Lin and Yifu Sun and Shirui Huang and Lin Niu and Gu Gong and Guojian Xiao and Bojian Zheng and Xiang Yuan and Qi Chen and Jie Xiao and Dongyang Zheng and Xiaofeng Yang and Kai Liu and Jianchen Zhu and Lifu Wang and Qinglin Lu and Jie Liu and Liang Dong and Fan Jiang and Ruibin Chen and Lei Wang and Chao Zhang and Jiaxin Lin and Hao Zhang and Zheng Ye and Peng He and Runzhou Wu and Yinhe Wu and Jiayao Du and Jupeng Chen and Xinyue Mao and Dongyuan Guo and Yixuan Tang and Yulin Tsai and Yonghao Tan and Jiaao Yu and Junlin Yu and Keren Zhang and Yifan Li and Peng Chen and Tian Liu and Di Wang and Yuhong Liu and Linus and Jie Jiang and Zhuo Chen and Chunchao Guo},
journal= {arXiv preprint arXiv:2509.12815},
year = {2025}
}
备注
Technical Report