MJ-Bench: Is Your Multimodal Reward Model Really a Good Judge for Text-to-Image Generation? | Read Paper on Bytez