1 paper · 1 filter
Yunze Tong, Mushui Liu, Canyu Zhao +7
Deploying GRPO on Flow Matching models has proven effective for text-to-image generation. However, existing paradigms typically propagate an outcome-based reward to all preceding d…