SIGGRAPH Asia 2025, HongKong
Yuxuan Xue1 , Xianghui Xie1, 2, Margaret Kostyrko1, Gerard Pons-Moll1, 2
1Real Virtual Human Group @ University of Tübingen & Tübingen AI Center
2Max Planck Institute for Informatics, Saarland Informatics Campus
- [2025/10/14] InfiniHuman paper is available on ArXiv.
- [2025/10/14] InfiniHumanData and InfiniHumanGen are scheduled to be released soon.
- Training 3D human generative models requires large-scale, diverse, and richly annotated datasets!
- Capturing and annotating real human data is prohibitively expensive and limited in scale and diversity!
- Can we distill foundation models to generate theoretically unbounded richly annotated 3D human data?
- InfiniHumanData: Automatic pipeline distilling vision-language and image generation models => 111K diverse identities with multi-granularity annotations
- Quality indistinguishable from real scans: Users cannot tell the difference between our synthetic data and real scan renderings!
- InfiniHumanGen: Diffusion-based generative model trained on InfiniHumanData => Fast, realistic, and precisely controllable 3D human generation from text, clothing, body shape, and pose
@article{xue2025infinihuman,
author = {Xue, Yuxuan and Xie, Xianghui and Kostyrko, Margaret and Pons-Moll, Gerard},
title = {InfiniHuman: Infinite 3D Human Creation with Precise Control},
booktitle = {SIGGRAPH Asia 2025 Conference Papers},
year = {2025},
}