3

Text-to-image diffusion models have achieved widespread popularity due to their unprecedented image generation capability. Their ability to synthesize and modify human faces has spurred research into using generated images in face recognition pipelines for robustness and bias mitigation. In this work, we identify their limitations and disparities in terms of image quality using a combination of qualitative and quantitative methods. Methodology and insights apply beyond face generation: similar analysis is applicable to other vision tasks, large language models, and audio generators.