4DAnyone open source: Turn casual phone video into consistent multi-view 4D avatars

aigclink · x · 2026-08-24

Open-sourced by Zhejiang University and Ant Research (SIGGRAPH Asia 2026), 4DAnyone turns casual monocular videos into multi-view videos for 4DGS reconstruction. The repo provides full installation and inference scripts with flexible camera configurations. Weights are downloaded automatically. FlashAttention-3 or SageAttention is recommended for faster inference.

Related event: Zhejiang University and Ant Group Open-Source 4DAnyone for Dynamic 3D Avatars from Phone Video(2 posts)→

Original post →

More from Multimodal

Multimodal channel →