AI-Faces by Illinois: A Database of AI-Generated Faces Depicting Children

Faces serve as gateways of impression formation during human interactions and are widely used in psychological research. To test these questions, researchers often rely on databases of photographs of adults’ and children’s faces for stimuli. Although the value of these resources is clear, there are some limitations, especially for research using children’s images as stimuli. This manuscript introduces a novel alternative to those databases: AI-Faces by Illinois, a fully open-access (CC-BY-NC-4.0) database of 1,152 photorealistic, AI-generated images depicting child faces. All images, norming data, and Online Supplemental Material are freely available on OSF (https://osf.io/vurm5/). These images vary systematically by perceived age (3, 6, 10, 15 years), gender (boy, girl), race/ethnicity (Black, East Asian, Latine, Middle Eastern/North African, Native American, Pacific Islander, South Asian, White), and emotional expression (10 smiling, 4 angry, 4 frowning per group). A norming study assessing 1,261 adults’ ratings of the images indicates that faces across race and gender groups are perceived as equally real and align with the gender and racial identities they were prompted to represent. Many well-documented perceptual biases observed with real faces—such as the Own-Race Bias—also generalized to the AI-Faces by Illinois images, providing evidence that similar processes are at play when adults evaluate AI-generated images of children. This manuscript highlights the methodological advantages, and new ethical considerations, of using AI-generated faces of children in psychological research.