1 paper · 1 filter
Zongyu Wu, Hongcheng Gao, Yueze Wang +2
Text-to-Image (T2I) models have shown great performance in generating images based on textual prompts. However, these models are vulnerable to unsafe input to generate unsafe conte…