Can Pre-Trained Text-to-Image Models Generate Visual Goals for Reinforcement Learning? | Read Paper on Bytez