Beauty-Naturalness Dissociation in AI-Edited Faces: An Eye-Tracking Study of Visual-Cognitive Mechanisms

With the growing use of AI-based facial editing and synthesis, understanding how people perceive the naturalness and realism of edited faces has become increasingly important. This study investigates that process in controlled AI-edited faces from a cognitive perspective. Using MediaPipe Face Mesh, we constructed a standardized facial stimulus pipeline and systematically manipulated four facial factors—eye size, mouth size, skin whitening, and face slimming—under a 2 × 2 × 2 × 2 within-subject design. Eye tracking and behavioral ratings were jointly collected to examine the dissociation between Beauty and Naturalness. The results revealed a clear separation between the two judgment dimensions: Beauty was more strongly associated with attention to the eyes and skin, whereas Naturalness was more strongly associated with attention to the facial contour, mouth, and nose. Based on these findings, we propose a dual-mechanism cognitive framework in which Beauty is linked to an attraction-oriented process and Naturalness is linked to a structure-oriented authenticity-checking process. These findings provide a visual-cognitive account of authenticity-related judgments in AI-edited faces.