Revolutionizing Text-to-Image Synthesis: UC Berkeley Researchers Utilize Large Language Models in a Two-Stage Generation Process for Enhanced Spatial and Common Sense Reasoning