Kooper-Johnson et al examine whether AI-generated images of skin conditions are realistic and accurate enough to be used for medical education. The researchers generated images of ten skin conditions using DALLE-2 and DALLE-3, randomized them with real clinical photographs, and asked dermatology residents and physicians to identify which images were AI-generated and to provide diagnoses. Performance was compared across image type, AI model, physician training level, and disease category. The results showed that AI-generated images were correctly identified as computer-generated in 70.8% of cases, with residents outperforming attendings, especially for DALLE-2 images. Diagnostic accuracy was significantly lower for AI-generated images than for clinical photographs (40.83% vs. 72%), and DALLE-3 images were diagnosed more accurately than DALLE-2 images. Attendings diagnosed DALLE-2 images more accurately than residents, while both groups performed similarly on DALLE-3 images. Infectious conditions were often flagged as AI-generated but had the lowest diagnostic accuracy, and skin of color was largely underrepresented.