Beyond One-Size-Fits-All: Personalizing Affective Stimuli with Generative AI

Abstract

Reliable affect induction remains a core challenge in affective science, as legacy stimulus databases are static, culturally limited, and often elicit only moderate emotional responses. This PhD project proposes a closed-loop pipeline leveraging modern AI models to generate, evaluate, explain, and personalize affective image stimuli. Generative models produce candidate images targeting specific emotions, while vision–language models predict their emotional impact and provide interpretable rationales. A reinforcement-learning agent then adaptively selects stimuli tailored to each individual. The central hypothesis is that personalized stimuli will elicit stronger target-emotion responses than standard, non-personalized images. Completed pilot studies demonstrate the feasibility of AI-based affect generation and prediction. The project aims to establish a transparent, scalable framework for personalized affect induction.