Weight bias is a widespread societal issue that materializes as discriminative behaviours, attitudes, and stereotypes towards people with obesity, and has seen a significant increase since the early 2000s, reaching levels of prevalence compared to those of racial discrimination. Implicit weight bias is particularly harmful, as it has been shown to threaten health outcomes of people with obesity both via internalization and lower quality healthcare. This study aimed to create and validate a set of AI-generated faces intended for use as stimuli in the Implicit Association Test (IAT) to assess implicit weight bias. Recognizing the lack of validated stimuli depicting individuals with obesity, we generated a diverse set of faces representing various ethnicities and ages. Fourty-eight stimuli were generated by AI and evaluated by a sample of 210 adults for realism, perceived origin (AI-generated vs. human-made), and several personality traits and affective features. Results demonstrated that the AI-generated faces were perceived as photorealistic, with participants nearly equally divided in perceiving them as real or AI-generated. Notably, the perceived weight of the stimuli was negatively correlated with attractiveness and competence, highlighting implicit negative stereotypes against individuals with obesity. Additionally, stimuli representing individuals with obesity were perceived as less realistic than those representing individuals of average weight. This validated set of stimuli fills a critical gap in research tools, offering robust, representative, and realistic images for studying implicit weight bias.