Diffusion Models Enhance Text-Rich Image Generation with Disentangled Learning

A novel framework named ARTIST that leverages disentangled learning and large language models to significantly improve the text rendering ability of diffusion models for generating high-quality text-rich images.