IEEE transactions on pattern analysis and machine intelligence
Refine, Control and Distill: A Text-to-Image Framework for Faithful Image Generation.
Peng Xing, Ning Wang, Yanpeng Sun, Jinhui Tang, Zechao Li
Published: 202510.1109/TPAMI.2025.3628109
Abstract
While text-to-image diffusion models exhibit outstanding results, they struggle to faithfully generate key subjects with corresponding attributes in prompts, challenges known as catastrophic neglect and attribute binding. Previous works typically uti…
Preview only. Read the full abstract at the source