No Evidence of Repeated Exposure Increasing the Perceived Realism of Real and AI-Generated Faces

Abstract

Processing fluency accounts and recent empirical evidence suggest that repeated exposure to images may increase their perceived realism. Here, focusing exclusively on faces, the present study tested whether repetition increased the likelihood that images were judged as real photographs rather than AI-generated. Participants first rated the perceived realism of a set containing both real photographs and synthetic faces. After completing a filler task, participants again rated the perceived realism of a subset of these images, intermixed with a set of new images. A funnel debriefing confirmed that participants were unaware of the study's predictions. Bayesian multilevel modelling found no evidence that repeated images were perceived as more real. Indeed, for AI-generated faces, repetition led to a small increase in perceived ‘synthetic-ness’. These results suggest that under conditions of minimal demand characteristics, graded responding and a single prior exposure, repeated viewing alone is insufficient to increase perceived realism for face images.