Computer Science · AIML
Zicheng Duan段子成
Video foundation models & generative world models
I am a final-year Ph.D. student in Computer Science at the Australian Institute for Machine Learning (AIML), University of Adelaide, supervised by Prof. Lingqiao Liu and Dr. Xinyu Zhang.
Before that, I completed my Master’s degree at the Australian National University, where I worked on multi-camera detection under the supervision of Prof. Liang Zheng.
My current research focuses on Video Foundation Models and Generative World Models. I am also a research intern at Alibaba Token Foundry, working closely with Dr. Xinlong Wang.
I am on the job market for Fall 2026 (expected to graduate in early 2027) — feel free to reach out!
我正在参加 2026 年秋招，预计于 2027 年上半年毕业，欢迎联系！
Research interests
I am currently interested in how visual generation can work with language-driven intelligence to create realistic, interactive virtual worlds, with my current research focusing on video world/foundation models, reinforcement learning for generative models, and code-to-anything.
Recent news
- Jun 2026LiveWorld is accepted to ECCV 2026.
- May 2026I have joined Alibaba Group - Alibaba Token Foundry as a full-time research intern, focusing on generative world modeling.
- Mar 2026
Earlier updates
- Feb 2026Three papers accepted to CVPR 2026.
- Nov 2025Invited talk at DICTA 2025 workshop on Generative World Models.
- Oct 2025One paper accepted to TMLR.
Selected research
All publicationsGenerative World Modeling
ECCV 2026
LiveWorld: Simulating Out-of-Sight Dynamics in Generative Video World Models
The first to formulate the decoupling of world state maintenance and observation rendering in generative video world models!
Controllable Content Generation
All publications 14
NeurIPS 2026 (Evaluations and Datasets Track)
TerraVis: Towards Evaluation of World-Grounded Visual Consistency in Text-to-Image Generation via MLLM Workflows
-
ECCV 2026
LiveWorld: Simulating Out-of-Sight Dynamics in Generative Video World Models
The first to formulate the decoupling of world state maintenance and observation rendering in generative video world models!
-
-
-
-
-
Internships
Alibaba Token Foundry
May 2026 – PresentBeijing, China
Mentor: Dr. Xinlong Wang
Topic: Video foundation models and video world models.
Alibaba DAMO Academy
Jan 2025 – Apr 2025Hangzhou, China
Mentor: Dr. Hao Luo
Topic: Multimodal image generation and editing.
Leonardo.AI / Canva
Feb 2024 – Jul 2024Sydney, Australia
Mentors: Mr. Ethan Smith, Prof. Martin Bell
Topic: Controllable image generation.
Institute of Automation, Chinese Academy of Sciences
Dec 2020 – Dec 2021Beijing, China
Mentors: Prof. Haoran Li, Prof. Dongbin Zhao
Topic: Multi-view 3D understanding systems for object detection.
Beyond the papers
Invited talk
Controllability Matters: Interpreting World Models from a Controllability Perspective
DICTA 2025 Workshop — Visual Generative Models: Past, Present, and Future
View slidesProject
The Chosen One
Unofficial implementation, selected as the official implementation by the author.
View code PaperCompetition
ICRA 2021 RoboMaster
ICRA 2021 Workshop RoboMaster Competition
Team Neurons · Ranked 6 / 26
Academic service
Reviewer CVPR’26, AAAI’25, DICTA’25, NeurIPS’24, ICASSP’24