Zehuan Huang (黄泽桓)

VAST

Email: huanngzh@gmail.com
[Github] [Google Scholar] [Twitter]

Biography

I am a researcher at VAST. I received my master's degree from Beihang University, supervised by Prof. Lu Sheng.

My research interest is generative models for interaction and simulation.

I am grateful to all my collaborators and mentors along the way. I first started doing research under the guidance of Prof. Miao Wang. Then I started working on deep learning related projects under the supervision of Prof. Lu Sheng. Besides, I also have interned at MiniMax, Shanghai AI Lab, VAST, and Tencent Hunyuan3D, and I'm fortunate to have worked closely with Junting Dong, Yuan-Chen Guo, Yanpei Cao, Yang Li, Zhuo Chen, and Chunchao Guo.

I am always open to academic and industrial collaborations, if you share the vision, please do not hesitate to contact me!

Educations

Sept. 2023 - Jan. 2026, Master, Beihang University, Beijing. Supervised by Prof. Lu Sheng. Shen Yuan Medal (highest honor for students in Beihang University).
Sept. 2019 - June 2023, Bachelor, Beihang University, Beijing. Outstanding Graduate Student of Beijing.

Industrial Experience

June 2025 - Jan. 2026, Tencent Hunyuan3D, Beijing, Top-Talent Research Intern.
Dec. 2023 - May 2025, VAST, Beijing, Research Intern.
Aug. 2023 - Dec. 2023, Shanghai Artificial Intelligence Laboratory, Beijing, Research Intern.
May 2022 - June 2023, MiniMax, Beijing, Intern.

News

[2026-10] EdenLive is now open-sourced, a step towards real-time interactive worlds.
[2025-11] One paper VoxHammer is accepted by 3DV 2026 as oral presentation.
[2025-06] One paper MV-Adapter is accepted by ICCV 2025.
[2025-03] Two papers MIDI-3D and Ouroboros3D are accepted by CVPR 2025.
[2024-03] Invited talk at AnySyn3D on Compositional 3D Scene Generation (video).

Selected Projects

EdenLive: Real-Time Generative Rendering from Any Visual Reference

Zehuan Huang*, Tian-Xing Xu*, Chunli Peng*, Cheng Zhang*, Yifei Song, Zeyi Zheng, Zhipeng Yu, Peng Wang†, Yan-Pei Cao^✉

VAST, 2026

[Project Page] / [Code] / [Weights]

TL;DR: Turns a proxy or reference video stream into photorealistic or stylized video in real time, at over 24 FPS.

Selected Publications

	SegviGen: Repurposing 3D Generative Model for Part Segmentation Lin Li, Haoran Feng, Zehuan Huang†, Haohua Chen, Wenbo Nie, Shaohua Hou, Keqing Fan, Pan Hu, Sheng Wang, Buyu Li, Lu Sheng^✉ SIGGRAPH 2026 [TOG] [Project Page] / [Paper] / [Code] TL;DR: Repurposes a native 3D generative model for interactive and full 3D part segmentation.
	Mira-Scene: Pixel-Aligned Layouts for Generative 3D Scene Reconstruction Yang-Tian Sun, Tianjia Liu, Zehuan Huang†, Yi-Hua Huang, Xiaoyang Lyu, Ziyi Yang, Zi-Xin Zou, Yuan-Chen Guo, Yan-Pei Cao^✉, Xiaojuan Qi^✉ Preprint [Project Page] / [Paper] / [Code] TL;DR: Recovers pixel-aligned layouts so a generative model can reconstruct an editable 3D scene from a single image.
	MV-Adapter: Multi-view Consistent Image Generation Made Easy Zehuan Huang, Yuan-Chen Guo, Haoran Wang, Ran Yi, Lizhuang Ma, Yan-Pei Cao^✉, Lu Sheng^✉ ICCV 2025 [Project Page] / [Paper] / [Code] / [机器之心] TL;DR: Versatile multi-view generation with various base models and conditions, and high-quality 3D texture generation.
	MIDI: Multi-Instance Diffusion for Single Image to 3D Scene Generation Zehuan Huang, Yuan-Chen Guo, Xingqiao An, Yunhan Yang, Yangguang Li, Zi-Xin Zou, Ding Liang, Xihui Liu, Yan-Pei Cao^✉, Lu Sheng^✉ CVPR 2025 [Project Page] / [Paper] / [Code] / [机器之心] TL;DR: MIDI-3D extends image-to-3D object generation models to multi-instance diffusion models for compositional 3D scene generation.
	VoxHammer: Training-Free Precise and Coherent 3D Editing in Native 3D Space Lin Li, Zehuan Huang**†, Haoran Feng, Gengxiong Zhuang, Rui Chen, Chunchao Guo, Lu Sheng^✉ 3DV 2026 Oral [Project Page] / [Paper] / [Code] TL;DR: A training-free 3D editing approach that performs precise and coherent editing in native 3D latent space instead of multi-view space.

	Stereo World Model: Camera-Guided Stereo Video Generation Yang-Tian Sun, Zehuan Huang, Yifan Niu, Lin Ma, Yan-Pei Cao, Yuewen Ma, Xiaojuan Qi CVPR 2026 [Project Page] / [Paper] / [Code] TL;DR: A camera-conditioned world model that generates stereo video with consistent appearance and binocular geometry.
	MultiLive: On-Policy Distillation of Video and Multi-View Priors for Multi-View Video Generation Yifan Niu, Yang-Tian Sun†, Haoran Feng, Lu Sheng, Yan-Pei Cao^✉, Zehuan Huang†^✉ Preprint [Project Page] TL;DR: Distills video and multi-view priors into one generator for synchronized multi-view video, without multi-view video training data.

* Equal contribution. † Project Lead. ✉ Corresponding author.

Selected Honors & Awards

[2025] Shen Yuan Medal (highest honor for students in Beihang University)
[2025] National Scholarship
[2025] Outstanding Graduate Student of Beihang University
[2024] National Scholarship
[2023] Outstanding Graduate Student of Beijing

	SegviGen: Repurposing 3D Generative Model for Part Segmentation Lin Li, Haoran Feng, Zehuan Huang†, Haohua Chen, Wenbo Nie, Shaohua Hou, Keqing Fan, Pan Hu, Sheng Wang, Buyu Li, Lu Sheng^✉ SIGGRAPH 2026 [TOG] [Project Page] / [Paper] / [Code] TL;DR: Repurposes a native 3D generative model for interactive and full 3D part segmentation.
	Mira-Scene: Pixel-Aligned Layouts for Generative 3D Scene Reconstruction Yang-Tian Sun, Tianjia Liu, Zehuan Huang†, Yi-Hua Huang, Xiaoyang Lyu, Ziyi Yang, Zi-Xin Zou, Yuan-Chen Guo, Yan-Pei Cao^✉, Xiaojuan Qi^✉ Preprint [Project Page] / [Paper] / [Code] TL;DR: Recovers pixel-aligned layouts so a generative model can reconstruct an editable 3D scene from a single image.
	MV-Adapter: Multi-view Consistent Image Generation Made Easy Zehuan Huang, Yuan-Chen Guo, Haoran Wang, Ran Yi, Lizhuang Ma, Yan-Pei Cao^✉, Lu Sheng^✉ ICCV 2025 [Project Page] / [Paper] / [Code] / [机器之心] TL;DR: Versatile multi-view generation with various base models and conditions, and high-quality 3D texture generation.
	MIDI: Multi-Instance Diffusion for Single Image to 3D Scene Generation Zehuan Huang, Yuan-Chen Guo, Xingqiao An, Yunhan Yang, Yangguang Li, Zi-Xin Zou, Ding Liang, Xihui Liu, Yan-Pei Cao^✉, Lu Sheng^✉ CVPR 2025 [Project Page] / [Paper] / [Code] / [机器之心] TL;DR: MIDI-3D extends image-to-3D object generation models to multi-instance diffusion models for compositional 3D scene generation.
	VoxHammer: Training-Free Precise and Coherent 3D Editing in Native 3D Space Lin Li, Zehuan Huang**†, Haoran Feng, Gengxiong Zhuang, Rui Chen, Chunchao Guo, Lu Sheng^✉ 3DV 2026 Oral [Project Page] / [Paper] / [Code] TL;DR: A training-free 3D editing approach that performs precise and coherent editing in native 3D latent space instead of multi-view space.