Programmed images and affective framing: analysing generative AI metadata in experiential visual communication

This study investigates how experiential contexts and affective categories are represented in the metadata accompanying artificial-intelligence (AI)-generated images across tourism, fashion, and retail. Using a consensus seeded topic model to identify industries in the AI usage suggestions of 23,278 images (tourism 3,957, fashion 3,325, retail 2,947) and community detection on the resulting word networks, the findings reveal industry-specific distributions of emotion labels and suggested reactions. Tourism-related entries were most frequently labelled with joy, admiration, and serenity; fashion entries with admiration, joy, and nostalgia; and retail entries with joy, excitement, and admiration. These patterns indicate that AI’s metadata reproduces industry-specific affective conventions rather than demonstrating that the system intentionally authors emotional experiences. The study conceptualises programmed images as image–metadata assemblages through which experiential contexts and affective categories are algorithmically assigned and demonstrates the utility of the valence–arousal model for organising AI-generated emotion labels.