DiSCO: Defending text-to-image generation through distribution-guided contrastive prompt optimization
Source
arxiv.org
Author
Tong Zhang, Motasem Alfarra, Carlos Hinojosa, Christos Louizos, Bernard Ghanem
Date
Why it matters
Identifies a class of prompts that pass text filters yet still produce unsafe images, and defends against it purely at the prompt layer — usable against closed image models where weight editing or encoder tuning is impossible.
articleBeyond Prompt Engineering: A Systematic Analysis of Prompt Lexical Sensitivity and Its Impacts on QualityQipeng Xie, Zi Liang, Jiafei Wu, Yufei Chen, Weizheng Wang, Wenao Ma, Zhong Ming, Haiqin Yang, Kaishun Wu
articleFew-Shot Degradation Is Not What It Seems: Behavioral Evidence, Representation Analysis, and a Random-Text Control Across 12 Models, 2 Tasks, and 2 ArchitecturesVolodymyr Ovcharov
articleSafety for Whom? Refusing the Right Subset of a Topic, Not the Whole Topichuggingface.co