Yunhan Yang Yufan Zhou Yuan-Chen Guo Zi-Xin Zou Yukun Huang Ying-Tian Liu Hao Xu Ding Liang Yan-Pei Cao Xihui Liu

Abstract

The creation of 3D assets with explicit, editable part structures is crucialfor advancing interactive applications, yet most generative methods produceonly monolithic shapes, limiting their utility. We introduce OmniPart, a novelframework for part-aware 3D object generation designed to achieve high semanticdecoupling among components while maintaining robust structural cohesion.OmniPart uniquely decouples this complex task into two synergistic stages: (1)an autoregressive structure planning module generates a controllable,variable-length sequence of 3D part bounding boxes, critically guided byflexible 2D part masks that allow for intuitive control over part decompositionwithout requiring direct correspondences or semantic labels; and (2) aspatially-conditioned rectified flow model, efficiently adapted from apre-trained holistic 3D generator, synthesizes all 3D parts simultaneously andconsistently within the planned layout. Our approach supports user-defined partgranularity, precise localization, and enables diverse downstream applications.Extensive experiments demonstrate that OmniPart achieves state-of-the-artperformance, paving the way for more interpretable, editable, and versatile 3Dcontent.