Reinforcement Learning Meets Masked Generative Models: Mask-GRPO for Text-to-Image Generation | Read Paper on Bytez