About Me

MPhil Student & AI Researcher

I'm Jianyu Lai (赖建宇), an MPhil student at HKUST(GZ) advised by Prof. Lei Zhu, focusing on the application of MLLM and AIGC. My first work SnowMaster (CVPR 2025) applies generative rewards and reinforcement learning to optimize image degradation assessment. I then co-developed PosterCraft (ICLR 2026) with Sixiang Chen, a poster generation model that achieves high aesthetic quality with accurate text rendering, gaining hundreds of GitHub stars. During my internship at Meituan, I completed PosterOmni (CVPR 2026) and PosterReward (CVPR 2026) for poster editing and reward modeling.
Jianyu Lai

Research Interests

Reward Modeling & MLLM. I am passionate about exploring MLLM applications in image evaluation and generating feedback signals. In SnowMaster [CVPR'25], we propose a method preference optimization and generative reward-based assessment methods to enhance MLLM perception of image degradation levels. In PosterReward [CVPR'26], I lead the development of a unified framework for training both discriminative and generative reward models.

Generative Models. I believe research should matter to real users. My contributions include PosterCraft [ICLR'26] for high-quality aesthetic poster generation and PosterOmni [CVPR'26] for versatile poster editing, pushing the boundaries of visual content creation and achieving widespread community adoption.

Education

  • Sept. 2024 – Jun. 2026: MPhil in Data Science and Analytics, The Hong Kong University of Science and Technology (Guangzhou)
  • Sept. 2020 – Jun. 2024: B.Eng. in Automation, South China University of Technology

Research Experience

  • Jan. 2026 – Present: Research Assistant, City University of Hong Kong
    Supervised by Prof. Kede Ma; Developing a multi-modal agent with spatial reasoning capabilities to guide aesthetic photography composition.
  • Jun. 2025 – Feb. 2026: Research Intern, Meituan, Intelligent Vision Department
    Supervised by Jialin Gao; Designing poster generation/editing models, vision reward models and the optimization of unified multimodal understanding and generation models.

Recent Achievements

PosterOmni accepted by CVPR'2026

PosterReward accepted by CVPR'2026

PosterCraft accepted by ICLR'2026

GenHaze accepted by ICCV'2025

SnowMaster accepted by CVPR'2025

📰 News and Updates

  • Feb. 2026: 🎓 PosterOmni has been accepted by CVPR'2026.
  • Feb. 2026: 🎓 PosterReward has been accepted by CVPR'2026.
  • Jan. 2026: 🎓 PosterCraft has been accepted by ICLR'2026.
  • Jun. 2025: 🚀 Announcing the release of PosterCraft! A new framework for high-quality aesthetic poster generation.
  • Jun. 2025: 🎉 Luckily, GenHaze has been accepted by ICCV'2025.
  • Mar. 2025: 🎉 Fortunately, SnowMaster has been accepted by CVPR'2025.
  • Sept. 2024: 🎓 Began my MPhil program in Data Science and Analytics at HKUST (Guangzhou).