Learning-based Text-to-Image (TTI) models like Stable Diffusion have
revolutionized the way visual content is generated in various domains. However,
recent research has shown that nonnegligible social bias exists in current
state-of-the-art TTI systems, which raises important concerns.