Using AI to Generate Affective Images: Methodology and Initial Library

This project aims to advance affective science by leveraging generative artificial intelligence (AI) to create a publicly accessible database of affect induction images: Library of AI-Generated Affective Images (LAI-GAI). Current limitations in image-based research include weak to moderate emotional elicitation effects, limited image diversity, and minimal cultural tailoring of images. First, we built a pipeline for utilizing generative AI to develop affective images that capitalized on existing datasets and emotion definitions. Second, using the pipeline, we developed a library of images (n = 480) and their descriptions across 12 discrete emotion categories. The project employed a human-in-the-loop approach to refine and validate these stimuli, ensuring their effectiveness and cultural relevance. We validated the library through three international studies involving 1,611 participants from 44 countries, who rated three types of images: (1) images from existing affective databases, (2) AI-generated images without cultural adjustments, and (3) AI-generated images adjusted to specific cultural contexts. The AI-generated images were as effective in eliciting affective responses as the images from existing affective databases. The culturally adjusted images were slightly more effective in eliciting targeted affective responses compared to their counterparts unadjusted to participants' cultural context. Furthermore, we calculated the smallest effect sizes of interest for affect induction research (d’s from 0.06 to 0.31). This work demonstrates that researchers can now generate high-quality affect induction stimuli cost-effectively and at scale, and tailor them to diverse contexts—overcoming longstanding barriers and laying the groundwork for future AI-driven methodologies in affective science.