Zhening Huang

I'm a final-year PhD student in Machine Learning at the University of Cambridge, supervised by Professor Joan Lasenby. My current research mostly focuses on two directions: using coding agents for 3D modeling (LiteReality, LiteReality-Agent, Articraft), and video world models (SpaceTimePilot).

During my PhD I spent time at Adobe Research, working with Chun-Hao Huang.

I am currently interning at Morpheus AI (Roblox) on video world models, mentored by Xun Huang and Yicong Hong.

I am actively looking for full-time positions starting January 2027 in robotics, simulation, and video world models. Whether you are hiring or just want to talk research in these areas, feel free to reach out.

News

[Oct 2026] My PhD thesis has been awarded runner-up for the BMVA Sullivan Doctoral Thesis Prize!
[Aug 2026] Our latest project is out: LiteReality-Agent, together with the LiteReality scanner app! Scan a room with the iOS app and the agent turns it into a complete, graphics-ready scene with articulated assets. Read the blog post for the full story, or try the interactive viewer and the App Store listing.
[July 2026] Carrying on my internship at Roblox, following Morpheus AI joining the company.
[May 2026] The Cam×Time dataset from SpaceTimePilot is out! Check out the dataset page.
[May 2026] Articraft is out everywhere! The technical report, the code, and the Articraft-10K dataset are all available. Articraft is an agentic system that writes programs to generate articulated 3D assets at scale.

More news

[April 2026] Started an internship at Morpheus AI, working on video world models with Xun Huang and Yicong Hong.
[Feb 2026] SpaceTimePilot is accepted at CVPR 2026! Huge thanks to all the co-authors! We are working on releasing the codebase and dataset.
[Jan 2026] We released the codebase for LiteReality! You can use it to scan your room and convert it to a compact, graphics-ready reconstruction with full PBR materials. Check out the code and examples.
[Dec 2025] We released SpaceTimePilot. Check out the video and webpage. SpaceTimePilot disentangles space and time in video diffusion model, for implicit 4D reconstruction and exploration.
[Oct 2025] Fortunate to receive the NeurIPS 2025 Scholar Award! Grateful for the support.
[Sep 2025] LiteReality is accepted at NeurIPS 2025! Thanks to all the collaborators.
[July 2025] We released LiteReality. Check out the video and webpage. Working on producing a good open source codebase for this!
[June 2025] Started internship at Adobe Research, working with Chun-Hao Huang. Amazing experience so far!
[July 2024] OpenIns3D is accepted at ECCV 2024. Thanks to all the collaborators.
[April 2024] Attended BMVA symposium on Multimodal Learning in London. OpenIns3D got the Best Paper Award!
[Sep 2023] We released OpenIns3D, the first 2D-input-free pipeline for open-world 3D instance segmentation.
[Jan 2023] Started a 3-month internship at Toshiba Cambridge AI Lab, working on language interaction with point clouds.
[May 2022] Very lucky to be the 2022 Girton Postgraduate Research Award recipient!

First / Co-first Author Publications

All publications →

LiteReality-Agent: An Agentic System for Interactable 3D Indoor Scene Reconstruction
Zhening Huang*, Yueyan Li*, Johnathan Chiu*, Xiaoyang Lyu, Matt Zhou, Yuxin Yao, Joan Lasenby, Shangzhe Wu
* Equal contribution

arXiv, 2026

Project Page / Paper / Viewer / Scanner App / GitHub

Read the blog →

TLDR: LiteReality-Agent is an open-source, end-to-end toolkit for reconstructing simulation-ready rooms. Scan a room with the iOS LiDAR app, and the agent authors the room as a Python program, producing a simulation-ready room with articulated assets in a simulation engine.

Articraft: An Agentic System for Scalable Articulated 3D Asset Generation
Matt Zhou*, Ruining Li*, Xiaoyang Lyu*, Zhaomou Song*, Zhening Huang*, Chuanxia Zheng, Christian Rupprecht, Andrea Vedaldi, Shangzhe Wu
* Equal contribution

NeurIPS 2026

Project Page / Paper / GitHub

TLDR: Articraft reduces articulated 3D asset generation to writing a program that builds the asset, and uses an LLM agent with a domain-specific SDK and validation harness to write those programs automatically.

SpaceTimePilot: Generative Rendering of Dynamic Scenes Across Space and Time
Zhening Huang, Hyeonho Jeong, Xuelin Chen, Yulia Gryaditskaya, Tuanfeng Y. Wang, Joan Lasenby, Chun-Hao Huang

CVPR 2026

Cam×Time dataset is available →

Project Page / Paper / Cam×Time Dataset / Video / GitHub

TLDR: SpaceTimePilot disentangles space and time in a video diffusion model for controllable generative rendering. Given a single input video of a dynamic scene, it steers both the camera viewpoint and the temporal motion within the scene, enabling free exploration across the 4D space–time domain.

LiteReality: Graphics-Ready 3D Scene Reconstruction from RGB-D Scans
Zhening Huang, Xiaoyang Wu, Fangcheng Zhong, Hengshuang Zhao, Matthias Nießner, Joan Lasenby

NeurIPS 2025

Workshop Spotlight at the 2nd Workshop on 3D-LLM/VLA, CVPR 2026

Project Page / Paper / Video / GitHub

TLDR: LiteReality is an automatic pipeline that converts RGB-D scans of indoor environments into graphics-ready scenes with high-quality meshes, PBR materials, and articulated objects ready for rendering and physics-based interactions.

OpenIns3D: Snap and Lookup for 3D Open-vocabulary Instance Segmentation
Zhening Huang, Xiaoyang Wu, Xi Chen, Hengshuang Zhao, Lei Zhu, Joan Lasenby

ECCV 2024

Best Paper Award at the BMVA Symposium on Multimodal Learning 2024

Project Page / Paper / Video / GitHub

TLDR: OpenIns3D proposes a "mask-snap-lookup" scheme to achieve 2D-input-free 3D open-world scene understanding, which attains SOTA performance across datasets, even with fewer input prerequisites.

Service

Reviewer for CVPR, ICCV, ECCV, NeurIPS, ICML, ICLR, SIGGRAPH, SIGGRAPH Asia, Eurographics, and 3DV.

Design and source code from Jon Barron's website