Abstract: We study a novel problem in this manuscript, that is, if a modern text-to-image diffusion model can tailor any image classifier across domains and categories. Existing domain adaption works ...