Part-X-MLLM: Part-aware 3D Multimodal Large Language Model

Chunshi Wang^1,2†, Junliang Ye^2,3†‡, Yunhan Yang^2,4†, Yang Li²,
Zizhuo Lin¹, Jun Zhu³, Zhuo Chen², Yawei Luo^1✉, Chunchao Guo^2✉
†Equal Contribution. ^‡Project Lead. ^✉Corresponding Authors.
¹Zhejiang University, ²Tencent Hunyuan, ³Tsinghua University, ⁴The University of Hong Kong

         

---

Abstract

We introduce Part-X-MLLM, a native 3D multimodal large language model that unifies diverse 3D tasks by formulating them as programs in a structured, executable grammar. Given an RGB point cloud and a natural language prompt, our model autoregressively generates a single, coherent token sequence encoding part-level bounding boxes, semantic descriptions, and edit commands. This structured output serves as a versatile interface to drive downstream geometry-aware modules for part-based generation and editing. By decoupling the symbolic planning from the geometric synthesis, our approach allows any compatible geometry engine to be controlled through a single, language-native frontend. We pre-train a dual-encoder architecture to disentangle structure from semantics and instruction-tune the model on a large-scale, part-centric dataset. Experiments demonstrate that our model excels at producing high-quality, structured plans, enabling state-of-the-art performance in grounded Q&A, compositional generation, and localized editing through one unified interface.

Citation

@inproceedings{wang2026part,
  title={Part-x-mllm: Part-aware 3d multimodal large language model},
  author={Wang, Chunshi and Ye, Junliang and Yang, Yunhan and Li, Yang and Lin, Zizhuo and Zhu, Jun and Chen, Zhuo and Luo, Yawei and Guo, Chunchao},
  booktitle={International Conference on Learning Representations},
  volume={2026},
  pages={150386--150417},
  year={2026}
}
@article{ye2026hunyuan3d,
  title={Hunyuan3D-Buffalo 1.0: A Unified Multimodal Model for Scalable 3D Generation, Understanding, and Editing},
  author={Ye, Junliang and Liu, Kenkun and Wang, Guocun and Li, Yang and Qu, Yansong and Wang, Chunshi and Xu, Jingwei and Yang, Yunhan and Zhao, Zibo and Xu, Jiachen and others},
  journal={arXiv preprint arXiv:2608.02711},
  year={2026}
}