AnyCrowd: Instance-Isolated Identity-Pose Binding for Arbitrary Multi-Character Animation

Zhenyu Xie, Ji Xia, Michael Christian Kampffmeyer, Panwen Hu, Zehua Ma, Yujian Zheng, Jing Wang, Zheng Chong, xujie zhang, Xianhang Cheng, Xiaodan Liang, Hao Li · 2026

Read AnyCrowd: Instance-Isolated Identity-Pose Binding for Arbitrary Multi-Character Animation by Zhenyu Xie, Ji Xia, Michael Christian Kampffmeyer, Panwen Hu, Zehua Ma, Yujian Zheng, Jing Wang, Zheng Chong, xujie zhang, Xianhang Cheng, Xiaodan Liang, Hao Li (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-15415.txt.

Description

Xie and co-authors present AnyCrowd, a video generation framework that animates many characters without mixing up their identities. Crowds often suffer identity bleeding, where faces and traits swap between people. Each character is encoded alone in its own latent block before the diffusion transformer runs. A three-stage attention scheme then binds each identity to its driving pose. A gated fusion module resolves clashes where characters overlap. The design scales to any crowd size while keeping each person's look stable across the clip.

Catalog record

CREATORZhenyu Xie, Ji Xia, Michael Christian Kampffmeyer, Panwen Hu, Zehua Ma, Yujian Zheng, Jing Wang, Zheng Chong, xujie zhang, Xianhang Cheng, Xiaodan Liang, Hao Li
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE64079
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.15415

Browse the full library · About the Tower of Babel