This research introduces a framework for comparative evaluation of human-curated versus AI-generated affective images using a multimodal AI agent. The dataset (N=80 pictures) includes a selection of 40 human-curated images from the Open Affective Standardized Image Set (OASIS), and a set of 40 synthetic images generated specifically for this study. The synthetic dataset was created by prompting the “GPT Image 1” model, a specialized image generation model built on GPT-4o, with the goal to represent four target emotional states—Excitement, Frustration, Boredom, and Relaxation. A custom AI agent was deployed to rate all images along the valence and arousal dimensions of the affective circumplex model. Statistical analyses were performed to compare: (1) human vs agent image ratings for OASIS and (2) the agent’s ratings of the AI-generated image set and OASIS. The findings indicate that the AI agent reliably aligned with the human ratings and that GPT-4o can serve as both a generator and evaluator of affective content, thus supporting scalable, human-free validation pipelines. This approach contributes to the field of affective computing by enabling rapid generation and analysis of emotionevoking stimuli, with potential applications in experimental psychology and mental health.