TL;DR: We distill foundation models to generate theoretically unbounded, richly annotated 3D human data with 111K identities, enabling fast, realistic, and precisely controllable avatar generation from text descriptions, body shape, and clothing images.
Generating diverse and controllable 3D human avatars is challenging due to the prohibitive cost of capturing and annotating large-scale datasets.
We introduce InfiniHuman, a framework that distills foundation models to generate richly annotated human data with minimal cost and unlimited scalability.
Our InfiniHumanData pipeline automatically creates 111K identities with unprecedented diversity in ethnicity, age, and clothing styles, each annotated with text descriptions, multi-view images, and SMPL body parameters.
Based on this, InfiniHumanGen enables fast, realistic, and precisely controllable avatar generation conditioned on text, body shape, and clothing assets.
Taking text description, explicit SMPL shape, and a cloth image as input, Gen-Schnell generates 3D-GS end-to-end, while Gen-HRes generates high-resolution textured mesh, both matched to input conditions.
InfiniHuman enables diverse and controllable 3D human generation from multiple input modalities. Below we showcase results from four different generation tasks: Generation from Text Prompts, Generation from Clothing, Generation from Body Pose & Shape, and Editing from Text. Click on any link to jump directly to that section, or scroll down to explore all results.
Watch our 5-minute voice-over explanation to understand how InfiniHuman works. This video provides a comprehensive overview of our approach, from data generation to the final controllable 3D human creation pipeline.
We appreciate our friends and colleagues in RVH for their feedback to improve the work. This work is made possible by funding from the Carl Zeiss Foundation. This work is also funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) - 409792180 (EmmyNoether Programme, project: Real Virtual Humans) and the German Federal Ministry of Education and Research (BMBF): Tübingen AI Center, FKZ: 01IS18039A. The authors thank the International Max Planck Research School for Intelligent Systems (IMPRS-IS) for supporting Y.Xue. G. Pons-Moll is a member of the Machine Learning Cluster of Excellence, EXC number 2064/1 – Project number 390727645.
@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},
}