A Prompt Engineering Method for Generating Emotional Images for Psychological Research

Standardized image datasets play an essential role in emotion research, but conventional creation methods often demand extensive manual effort and are thereby limited in scope. Recent advancements in Artificial Intelligence (AI), especially the Text-to-Image Synthesis (TIS) technique combined with prompt engineering, could offer a promising alternative for generating emotional images for psychological research. The present study introduces a novel procedure for emotional image generation, consisting of Seed Prompt Formation, Prompt Set Generation, and TIS to generate emotional images. Using Midjourney and ChatGPT as tools, we evaluated the effectiveness of this procedure through the creation of an AI-Generated Emotional Image Set (AGEIS), which encompasses seven emotion categories (sadness, disgust, amusement, inspiration, fear, tenderness, and neutral). Human evaluations revealed that the AI-generated images in AGEIS induced target emotions effectively and specifically. Additionally, AGEIS images in general were perceived as comparably authentic to real photos in terms of their likelihood of being AI-generated, although AGEIS images depicting disgust and tenderness were more likely to be identified as AI-generated. Taken together, this procedure offers a cost-effective and scalable method for emotional image generation, potentially advancing emotion research into new directions such as large-scale and longitudinal studies.