What makes a face destined for fame? A study published February 12, 2025, in the Journal of Marketing Research offers a possible answer, using artificial intelligence to predict celebrity visual potential – termed charisma – from human faces.
Led by Xiaohang Feng at Carnegie Mellon University, the research team developed a deep learning model that analyzes facial features to distinguish celebrities from non-celebrities with 95.92% accuracy.
The findings reveal that features like large eyes and sexual dimorphism enhance charisma, while a wide face or babyfaceness detract from it, providing insights for industries from entertainment to business.
Decoding Charisma Through AI
Charisma, defined as “a personality trait that differentiates someone from ordinary people” with “exceptional powers and qualities that elicit devotion from others,” is a coveted quality in media, politics, and leadership.
While past research focused on personality traits like dominance or warmth, this study pioneers the use of facial features to predict charisma. “
We hypothesize that facial features can predict charisma – a topic that is not yet studied,” the researchers write.
The researchers used a computer program to study 12,000 photos, including 6,000 of real celebrities like actors and singers, to learn what makes a face look famous. They then gave each new face a charisma score from 0 to 1 based on how closely its features match those of celebrities.
The study identified 11 facial features that play a role.
From most to least influential on charisma, those facial features were ranked as follows: facial width-to-height ratio, sexual dimorphism (distinctly masculine or feminine traits), averageness (how typical a face looks compared to others), high cheekbones, skin color, thin jaw, mouth-chin distance, large eyes, symmetry, mouth-nose distance, and babyfaceness (baby-like features such as large eyes and small chin).
Unlike attractiveness, which focuses on physical beauty like symmetry or clear skin, the charisma score measures a face’s star quality – its ability to seem influential or memorable, like a celebrity’s, based on features tied to traits like dominance or warmth.
Surprisingly, “a negative correlation between averageness and charisma” contradicts prior theories that average faces are more attractive, suggesting distinctive faces stand out more.
A Robust Three-Step Approach
They first used a dataset of 722,418 celebrity and 158,611 non-celebrity images, selecting 12,000 for training and 10,000 for testing.
The next step – interpretation – ranked the features’ contributions, revealing facial width-to-height ratio, sexual dimorphism, and averageness as top influencers.
They then compared their empirical findings with theoretical predictions, clarifying how personality traits like trustworthiness mediate these effects.
The study validated its model in different contexts. On Instagram, for example, they found that influencers whose selfies had higher charisma scores also had more followers.
Likewise, LinkedIn profiles showed C-suite executives had higher charisma scores than average employees.
The dataset was diverse, with celebrity images (29.43% female, 65.06% White, 82.28% aged 20-40) and non-celebrity images (29% female, 56% White, 94% aged 20-40) from nine benchmark face recognition datasets.
Implications for Fame and Beyond
“The charisma score may be a useful tool for screening human photos that are intended to sway an audience (e.g., in advertisements),” the study notes. Talent agencies could optimize clients’ appearances, while businesses might screen candidates for leadership roles.
The model’s generalizability across Instagram and LinkedIn suggests broad applicability, though the researchers caution about ethical concerns, such as bias in facial recognition.
By revealing that distinctive, non-average faces drive charisma, the study challenges conventional beauty standards and offers a new lens for spotting star potential.
Study Details:
- Publication: Journal of Marketing Research, American Marketing Association
- Publication Date: February 12, 2025
- Title: “EXPRESS: An AI Method to Score Celebrity Visual Potential”
- Authors: Xiaohang Feng, Shunyuan Zhang, Xiao Liu, Kannan Srinivasan, Cait Lamberton
- Affiliations:
- Tepper School of Business, Carnegie Mellon University (Feng, Srinivasan)
- Harvard Business School, Harvard University (Zhang)
- Stern School of Business, New York University (Liu)
- Wharton School, University of Pennsylvania (Lamberton)
- DOI: 10.1177/002224372513232