I am a 4th-year PhD student in Electrical and Computer Engineering (ECE) at The University of Hong Kong (HKU-MMLAB), advised by Prof. Xihui Liu. My research focuses on multimodal generative AI and agent systems, especially video world models, cinematic video generation, and controllable visual content generation. Previously, I obtained my bachelor's and master's degrees at Shanghai Jiao Tong University and a double master's degree (Diplome d'Ingenieur, General Engineering) at Ecole Centrale Paris, and conducted research internships at Tencent IEG ARC, Kuaishou Kling, and Huawei Noah's Ark Lab.
My work spans three areas: (1) video world models (Multics and MemLearner); (2) cinematic video and film generation (CineScene and FilMaster); and (3) compositional image and video generation (T2I-CompBench, T2I-CompBench++, T2V-CompBench, and GenMAC). The T2I-CompBench series provides widely used benchmarks for compositional text-to-image generation and has been adopted in major industrial model evaluations.
I am on the job market (graduating in 2027). If you are interested in our research, feel free to contact me via email.
Kaiyi Huang, Wangbo Yu, Yukun Huang, Jiwen Yu, Kunhao Liu, Donglin Yang, Wenbo Hu, Ying Shan, Xihui Liu
Preprint 2026
A multi-player world model with hierarchical communication for cross-player consistency and player-specific control, scaling from 4 to 10 players without retraining.
Kaiyi Huang, Yukun Huang, Yu Li, Jianhong Bai, Xintao Wang, Zinan Lin, Xuefei Ning, Jiwen Yu, Pengfei Wan, Yu Wang, Xihui Liu
CVPR 2026Highlight, top 2.8%
Scene-consistent cinematic video generation with implicit 3D-aware representations.
Kaiyi Huang, Yukun Huang, Xintao Wang, Zinan Lin, Xuefei Ning, Pengfei Wan, Di Zhang, Yu Wang, Xihui Liu
ICLR 2026
An end-to-end system that brings camera-language design and cinematic-rhythm control into automated film generation.
Kaiyi Huang, Yukun Huang, Xuefei Ning, Zinan Lin, Yu Wang, Xihui Liu
AAAI 2026
An iterative, self-correcting multi-agent framework for compositional text-to-video generation.
Kaiyi Huang, Chengqi Duan, Kaiyue Sun, Enze Xie, Zhenguo Li, Xihui Liu
TPAMI 2025Top-tier SCI Journal · IF 20.4 (2025)
An enhanced benchmark with broader compositional categories and improved evaluation metrics.
Kaiyi Huang, Kaiyue Sun, Enze Xie, Zhenguo Li, Xihui Liu
NeurIPS 2023
Adopted as standard evaluation metrics by DALL·E 3 (OpenAI), Stable Diffusion 3 (Stability AI), and PixArt-alpha (Huawei).
Kaiyue Sun, Kaiyi Huang, Xian Liu, Yue Wu, Zihan Xu, Zhenguo Li, Xihui Liu
CVPR 2025
A benchmark and evaluation suite for compositional text-to-video generation.
Co-organizing the 1st Workshop on Video World Models at CVPR 2026. One paper (MemLearner) was accepted to ECCV 2026.
Started a research internship at Tencent IEG ARC.
CineScene was accepted to CVPR 2026 as a Highlight.
FilMaster was accepted to ICLR 2026.
GenMAC was accepted to AAAI 2026.
One paper (DreamCube) was accepted to ICCV 2025.
One paper (T2V-CompBench) was accepted to CVPR 2025.
Started a research internship at Kuaishou Kling.
T2I-CompBench++ was accepted to TPAMI.
T2I-CompBench was adopted in the evaluation of DALL·E 3 (OpenAI).
T2I-CompBench was accepted to the NeurIPS 2023 Datasets and Benchmarks Track.
Began PhD study at The University of Hong Kong (HKU-MMLAB).
