Intermediate Text Representation Guided Text-to-Image Generation for Enhancing One-and-Only Alignment

Soyoun Won, Aryan Yazdan Parast, Basim Azam +2

Text-to-image (T2I) diffusion models often fail to faithfully render explicit textual descriptions, instead defaulting to strongly learned visual priors due to a phenomenon referre…