AI-generated food stimuli match real counterparts in perceived healthiness and calorie content but not in realism or willingness to eat

Abstract

Abstract Visual food stimuli are widely used in food perception and evaluation research, but are often limited by a lack of control of confounding variables. AI-generated food images may be a viable but yet understudied alternative. In a within-subject online experiment, 87 participants rated 60 pairs of AI-generated and pre-validated real visual food stimuli, balanced for calorie content and flavor, on four 100ms visual analogue scales (realism, willingness to eat, calorie content, healthiness). AI-generated visual food stimuli were rated as significantly less realistic and elicited lower willingness to eat compared to real visual food stimuli. Ratings of healthiness and calorie content did not differ between real and AI-generated stimuli. The perceived realism of AI-generated visual food stimuli relative to their real counterparts predicted relative ratings of healthiness and willingness to eat, indicating that AI hyperrealism is associated with higher secondary food-related evaluations. The results show that AI-generated visual food stimuli are perceived similar to real counterparts in aspects of food quality evaluation (healthiness, calorie content), yet not regarding palatability (realism, willingness to eat). Taken together, although AI-generated visual food stimuli can be associated with enhanced food evaluations beyond those of real food, they are perceived as less realistic than their real counterparts.