Zeyu JIANG
PhD Student (Year 2), Department of Electrical Engineering, City University of Hong Kong.
zeyujiang4-c@my.cityu.edu.hk
83 Tat Chee Avenue, Kowloon Tong, Kowloon, Hong Kong
I am a Ph.D. student in the Department of Electrical Engineering at the City University of Hong Kong (CityUHK).
My research lies in generative modeling and multimodal learning, with a primary focus on advancing the controllability, compositionality, and alignment of modern visual generative systems.
My recent work centers around Diffusion-based architectures, particularly in scenarios where generation is guided by structured conditions such as multiple reference images or iterative feedback. I am interested in moving beyond stateless generation paradigms toward experience-aware and structure-preserving generation frameworks, enabling models to better retain consistency and adapt across complex tasks.
My current research directions include:
- controllable and compositional image synthesis
- diffusion and transformer-based generative models
- multi-reference and multi-image editing
- preference learning and alignment for generative models (e.g., DPO, RL-based methods)
- representation learning for structure consistency and semantic disentanglement
I have been involved in several research projects on image generation and editing, including works on multi-reference image editing, region-aware adaptation, and copyright-aware generation. My recent papers have been accepted by venues such as CVPR and AAIML 2026, covering topics ranging from generalized cross-image editing to alignment-aware generation. More broadly, I am interested in understanding how generative models can develop persistent knowledge, controllable behaviors, and reduced “AI artifacts”, especially in long-horizon or user-interactive generation settings.