CLIP in Mirror: Disentangling text from visual images through reflection | Read Paper on Bytez