Developed an end-to-end system for text-to-image generation using prompt refinement and Stable Diffusion 2.1. Improved prompt quality 3× (ROUGE metrics) via T5-Small, with an average prompt-image alignment of 0.72