Hui Li 李慧

I am a Ph.D. student in Computer Science at Fudan University. My research focuses on visual generative models, especially diffusion and flow-based image/video generation, human-centric video generation, and efficient generative model training and inference.

I am advised by Prof. Siyu Zhu (朱思语) and Dr. Jingdong Wang (王井东).

Email Google Scholar GitHub CV
Computer Vision Generative AI Diffusion & Flow Models
Hui Li profile photo

Research

Visual Generation

Diffusion models, flow matching, image generation, and high-quality controllable synthesis.

Human-Centric Video

Large-scale video datasets, motion-aware filtering, portrait animation, and talking video generation.

Efficient Modeling

Pyramidal patchification, accelerated inference, and reducing train-test discrepancy in generative models.

Selected Work

Projects & Publications

OpenHumanVid paper pipeline figure
Human Video

OpenHumanVid: A Large-Scale High-Quality Dataset for Enhancing Human-Centric Video Generation

Hui Li*, Mingwang Xu*, Yun Zhan, Shan Mu, Jiaye Li, Kaihui Cheng, Yuxuan Chen, Tan Chen, Mao Ye, Jingdong Wang, Siyu Zhu

CVPR 2025 Highlight

A large-scale human-centric video dataset with precise captions and motion conditions, plus a filtering pipeline for improving video diffusion model training.

Paper GitHub · 346 stars
PPFlow paper teaser figure
Flow Models

Pyramidal Patchification Flow for Visual Generation

Hui Li, Baoyou Chen, Liwei Zhang, Jiaye Li, Jingdong Wang, Siyu Zhu

ICLR 2026

A pyramidal patchification strategy for diffusion and flow-based models that improves generation efficiency while preserving or improving visual quality.

Paper GitHub
MixFlow paper method figure
Training

MixFlow Training: Alleviating Exposure Bias with Slowed Interpolation Mixture

Hui Li, Jiayue Lyu, Fuyun Wang, Kaihui Cheng, Siyu Zhu, Jingdong Wang

CVPR 2026

A training method that mitigates train-test discrepancy in diffusion and flow-based generation by mixing slowed interpolation states into model training.

Paper GitHub
Hallo3 paper teaser image
Portrait Animation

Hallo3: Highly Dynamic and Realistic Portrait Image Animation with Video Diffusion Transformer

Jiahao Cui, Hui Li, Yun Zhan, Hanlin Shang, Kaihui Cheng, Yuqi Ma, Shan Mu, Hang Zhou, Jingdong Wang, Siyu Zhu

CVPR 2025

A portrait animation framework built on pretrained video diffusion transformers for more dynamic, realistic talking videos with natural backgrounds and motion.

Paper GitHub · 1,385 stars
Hallo2 paper teaser image
Portrait Animation

Hallo2: Long-Duration and High-Resolution Audio-Driven Portrait Image Animation

Jiahao Cui*, Hui Li*, Yao Yao, Hao Zhu, Hanlin Shang, Kaihui Cheng, Hang Zhou, Siyu Zhu, Jingdong Wang

ICLR 2025

A long-duration portrait animation system that extends audio-driven generation to high-resolution, temporally consistent videos.

Paper GitHub · 3,701 stars
Hallo paper visual result figure
Portrait Animation

Hallo: Hierarchical Audio-Driven Visual Synthesis for Portrait Image Animation

Mingwang Xu*, Hui Li*, Qingkun Su*, Hanlin Shang, Liwei Zhang, Ce Liu, Jingdong Wang, Yao Yao, Siyu Zhu

arXiv 2024

An open-source diffusion-based portrait animation framework with hierarchical audio-driven visual synthesis for lip, expression, and pose control.

Paper GitHub · 8,646 stars

Academic Service

Reviewer

Background

Education & Experience

2024 - 2028

Fudan University

Ph.D. in Computer Science and Technology, School of Computer Science

2020 - 2023

Jilin University

M.S. in Mathematics, College of Artificial Intelligence

2016 - 2020

Jilin University

B.S. in Information and Computing Science, School of Mathematics

2023

Zhejiang Lab

Algorithm Engineer. Worked on ControlGIF for image-to-video generation and video editing with AnimateDiff and ControlNet.