PS70 Artificial intelligence-generated faces in dermatological perception research: effects on perceived severity, realism and attractiveness of visible skin conditions

Abstract

Inflammatory dermatoses, particularly when affecting the face, may be associated with psychological and social burden, resulting in stigma, avoidance and reduced quality of life. Public perceptions of skin disease are influenced by visual cues, and artificial intelligence (AI)-generated faces are increasingly adopted in perception research. However, it remains unclear whether AI-generated facial images depicting skin diseases are perceived in the same way as real images. The present experiment examined whether image type and the presence of visible skin conditions influence the perception of faces. Participants (n = 141; general public sample) completed an online experiment using a 2 × 2 ­within-participants design manipulating image type (AI-generated vs. real) and visible skin condition (present vs. absent). Participants rated real and AI-generated faces (generated using Midjourney AI software), presented with and without a visible skin condition, on perceived realism, skin condition severity and attractiveness. Data were analysed using linear mixed-effects models. Real faces were rated as more realistic than AI-generated faces (P < 0.001), and faces without a visible skin condition were rated as more realistic than those with a skin condition (P < 0.001). Importantly, the presence of a skin condition reduced perceived realism more strongly for AI-generated faces than for real faces (P < 0.001). Real faces were rated as having a worse skin condition (higher severity) than AI-generated faces (P < 0.001). Attractiveness ratings were higher for AI-generated faces and those without a visible skin condition (P < 0.001), with no interaction between image type and skin condition (P > 0.05). We concluded that AI-generated and real facial images are not perceived equivalently when depicting visible skin conditions. While AI-generated faces can successfully convey the presence of a skin condition, they differ from real images in perceived realism, severity and attractiveness. These findings highlight the need to validate synthetic facial stimuli before use in psychodermatology research and to understand psychological wellbeing in individuals with visible skin conditions.