Hello, I’m Zeyue Xue (薛泽岳), a researcher with a passion for building Generative AI products and platforms. I am also a PhD candidate at The University of Hong Kong (MMLAB@HKU), advised by Ping Luo and Wenping Wang. I received my Bachelor’s degree from Huazhong University of Science and Technology with a ranking top 1%. Recently, my research interests have been in building unified multimodal models like Transfusion.

Research Interests

• Large-scale deep learning
• Multimodal generation

Three Representative Works

1. Zeyue Xue*, Guanglu Song*, Qiushan Guo, Boxiao Liu, Zhuofan Zong, Yu Liu, Ping Luo. “RAPHAEL: Text-to-Image Generation via Large Mixture of Diffusion Paths”, NeurIPS 2023. Model API (TL;DR-Training a diffusion foundation model via 1,000 NVIDIA A100s, the product model of SenseTime-SenseMirage.)
2. Zeyue Xue, Jianming Liang, Guanglu Song, Zhuofan Zong, Liang Chen, Yu Liu, Ping Luo. “Large-batch Optimization for Dense Visual Predictions: Training Faster R-CNN in 4.2 minutes”, NeurIPS 2022. Code (TL;DR-Training visual detectors via 768 NVIDIA A100s.)
3. Zeyue Xue, Jie Wu, Yu Gao, Fangyuan Kong, Lingting Zhu, Mengzhao Chen, Zhiheng Liu, Wei Liu, Qiushan Guo, Weilin Huang, Ping Luo. “DanceGRPO: Unleashing GRPO on Visual Generation”, Seed Tech Report, Code (TL;DR-The first unified RL-based framwork for visual generation.)

Invited Talks

2025-06: invited talks at PKU, ZJU, Xiaomi, Pika, Adobe, Anuttacon