1 paper
Yifan Yuan, Xiangyu Liu, Hongming Shan +5
Although text-to-image diffusion models generally exhibit strong prompt-following ability, we identify a persistent and previously underexplored failure pattern in which a small su…