Studies of emotion often rely on standardized stimulus sets to elicit affective responses. Although established databases provide images with normative valence and arousal ratings, selecting suitable stimuli can be difficult when experiments require specific thematic or content constraints. This challenge is especially pronounced for negative stimuli, which are central to research on maladaptive emotions and behaviors in clinical contexts but are often scarce in necessary quantity or specificity. The present study evaluated the feasibility of using generative AI, specifically text-to-image generators, to create tailored negative and neutral affective stimuli. To assess whether these images can serve as alternatives to traditional stimuli, we compared their affective properties to those reported in standardized image databases. Across two studies, participants rated the valence and arousal of 160 and 200 AI-generated images. Our findings revealed that AI-generated negative and neutral images reproduced the characteristic inverse association between valence and arousal observed in standardized databases, with moderate to strong correlations between these dimensions. These results highlight the potential of generative AI as a practical methodological tool for creating customized affective stimuli aligned with specific research objectives and experimental designs.
Chiu et al. (2026) studied this question.