Generative Models / Visual Intelligence

I work on scalable training and sampling methods of generative models.

I am a PhD student at Nanjing University, advised by Prof. Limin Wang. My research focuses on scalable training and inference for generative models, mainly diffusion models.

During my undergraduate study at Northwestern Polytechnical University, I was fortunate to learn from Prof. Yuchao Dai and Dr. Bo Li.

Representative works include DDT, whose decoupled diffusion transformer design has been used in follow-up systems such as RAE, PixelDiT, and PixNerd; PixNerd, the first practical large-patch diffusion transformer for pixel-space generation; and UniDDT, which validates scaling across different visual representation spaces and truly unifies multimodal understanding and generation.

Education & Honors

Nanjing University

Ph.D. (direct-track), Computer Science and Technology

  • National Scholarship (PhD)
  • Tencent Scholarship

Northwestern Polytechnical University

B.E., Computer Science and Technology

  • National Scholarship (Undergraduate)

Experience

ByteDance Seed

Research Intern, Seedream Group · large-scale rendered data, text editing

Alibaba Alimama

Research Intern · diffusion step distillation, inference sampling acceleration

SenseTime

Research Intern, autonomous driving perception

Selected Publications

Selected collaborations: Cubic Discrete Diffusion (CVPR 2026), Deco (CVPR 2026), MotionRAG (NeurIPS 2026), Flowing Backwards (AAAI 2026), DMM (2025). Full list on Google Scholar.