An AI Method to Score Celebrity Visual Potential from Human Faces

Celebrities have extraordinary abilities to attract and influence others. Predicting celebrity visual potential is important in the domains of business, politics, media, and entertainment. Can we use human faces to predict celebrity visual potential? If so, which facial features have the most impact on celebrity visual potential? We develop a three-step empirical framework that leverages computer vision techniques to predict celebrity visual potential from face images. In the prediction step, we optimize a ResNet-50 deep learning model on a large dataset of 6,000 celebrity images and 6,000 non-celebrity images and achieve 95.92% accuracy. In the interpretation step, we draw on psychology, economics, and behavioral marketing literature to select 11 interpretable facial features (e.g., width-to-height ratio). We calculate the direction and strength of the feature’s correlation with celebrity visual potential. We find that the facial width-to-height ratio, babyfaceness, and thin jaw contribute negatively to celebrity visual potential while sexual dimorphism, dark skin color, and large eyes contribute positively. In the mechanism step, we compare the interpretation results with extant theoretical relationships between facial features and celebrity visual potential, with personality traits as mediators. Contradicting theoretical predictions, we discover a negative correlation between averageness and celebrity visual potential. We demonstrate the generalizability of our results to media/entertainment and business domains. We also conduct experiments to compare our model-predicted scores with human-rated scores on celebrity-visual potential for further validation.