Pixel-Perfect: A Quantitative Analysis of Artificially Enhanced Selfies in a Young Adult Population
Artificial intelligence (AI) has rapidly integrated into daily life, influencing not only how individuals process information but also how they perceive and present their physical appearance. In recent years, AI’s ability to enhance facial aesthetics has garnered widespread attention, particularly with the rise of AI-enhanced selfies. This trend has been propelled by viral social media phenomena, such as the viral “AI headshot” trend on platforms like TikTok, where users engage with AI-powered applications to modify their selfies.
AI-powered photograph editing applications work by adjusting key facial features including skin texture, facial symmetry, and proportions. These modifications often aim to achieve more balanced facial structures and enhanced proportions, projecting a digitally perfected version of the user. The use of these AI-powered photograph editing applications extends beyond social media and has infiltrated clinical practice. Patients are increasingly bringing enhanced photographs to their consultations, using them as visual references for their desired aesthetic outcomes. Othman and colleagues1 found that face-editing applications influence patient decision-making to pursue cosmetic surgery.
To date, much of the focus within the scientific literature has been on the psychological impact of AI-enhanced images, with less attention given to the specific quantifiable changes these technologies make to facial features.2 This gap in research is significant because understanding the physical modifications made by AI tools can help providers set expectations between digital ideals and achievable results. This study aims to better understand the AI-driven modifications of facial features by quantifying changes in facial landmarks between original and AI-enhanced selfies.
Methods
Approval for the proposed study was granted by the Georgetown University Medical Center Institutional Review Board (577). Participants were recruited by university email invitations. The criteria for inclusion were age over 18 years and consenting to have photographs used for research purposes. All participants provided basic demographic information, including age, gender, and race. Participants submitted 9 selfies in standardized head positions: 4 with a dominant right profile, 4 with a dominant left profile, and 1 from a frontal view. Participants were instructed to smile, with their teeth visible, in all selfies. Next, each participant’s 9 selfies were uploaded to the Remini AI Photo Enhancer (Bending Spoons S.p.A.) smartphone application that is available for download on iOS and Android devices. The images were processed using the “Casual Headshot” pack, a feature designed to generate professional headshot-style enhancements. Within this pack, a standardized model (one for male and one for female participants) was selected as the reference for all image enhancements to ensure consistency across participants. This specific option was chosen to for all images to ensure consistency and standardization in the enhancement process because this mode provided a reproducible, professional aesthetic.
Each upload of 9 photographs generated 15 AI-enhanced images. The most frontal view image was selected for comparison with the original, nonenhanced image (Figure 1). Emotrics software was used to assess changes in facial measurements. This is an open-source machine-learning algorithm that allows for automated, objective measurements of facial landmarks (see Supplemental Digital Content 1, Figure 1, https://links.lww.com/DSS/B649). Any necessary adjustments were made by manually repositioning facial landmarks after image inspection by the research team to ensure measurement accuracy. Although Emotrics was originally developed for evaluating patients with facial paralysis, previous studies have demonstrated its application in analyzing user-driven, digitally manipulated photographs.3 Continuous variables were presented as mean values with standard deviations, whereas categorical variables were reported as frequencies and percentages. Paired 2-tailed t tests were conducted to compare facial measurements between original and enhanced selfies.
Figure 1.
Examples of frontal view original (A and C) and enhanced (B and D) selfies.
Source
Pixel-Perfect: A Quantitative Analysis of Artificially Enhanced Selfies in a Young Adult Population
Dermatologic Surgery : May 21, 2025
Results
A total of 20 participants (10 male and 10 female participants) were in included in the study. Most participants were between the ages of 25 to 34 years (90%).
For all participants, analysis revealed that AI significantly reduced brow height, commissure excursion, smile angle, and dental show while significantly increasing marginal reflex distance 2 (MRD2, Table 1). Dental show deviation, calculated as the difference between the right and left sides, was also found to be significantly decreased in the AI-enhanced image.
TABLE 1. – Comparison of Mean Facial Measurements for Original and Enhanced Images for all Subjects
When stratified by gender, commissure excursion, smile angle, and dental show remained significantly reduced for male and female participants (see Supplemental Digital Content 1, Table 1, https://links.lww.com/DSS/B649). In addition, AI significantly lowered brow height in male participant, although this was not found to be significant for female participants. In male participants, dental show deviation and smile angle deviation were significantly decreased. MRD2, both left and right, showed a significant increase in men. This trend was evident in female participants as well, although significance was only found on the left side.
Discussion
To the authors’ knowledge, this is the first study to quantify changes in AI-modified facial landmarks. In both men and women, AI-enhanced photographs showed a significant reduction in commissure excursion, smile angle, and dental show. When combined with the overall decrease in dental show deviation across all participants, this suggests that AI modifications may result in more subtle and symmetrical smiles. This aligns with previous research indicating that balance, rather than increased measurements, is crucial for an effective smile.4 Interestingly, this contrasts with prior research by Parsa and colleagues,3 which observed a tendency for female participants to increase their smile angle in digitally altered selfies. This difference can perhaps be explained by the fact that AI tends to optimize facial features based on aesthetic algorithms that prioritize symmetry, balance, and subtlety across a broad range of faces. By contrast, when individuals edit their own selfies, personal preferences and cultural/societal influences likely shape their modifications, adding a personal level of nuance and intention that AI-driven alterations lack.
This analysis revealed significantly reduced brow height in all participants. When stratified by gender, this decrease remained significant in men. This finding aligns with societal preferences for male facial features as a lower brow height is often associated with a more masculine or dominant appearance.3 The trend toward decreased brow height is in line with previous research showing that both men and women prefer lower brow heights in digitally altered selfies. Researchers have recently commented on the aesthetic shift in brow placement, particularly in female participants, as lower brows have been regarded as more attractive and youthful.3 Therefore, it follows that AI-generated enhanced images reflect these prevailing trends. Aesthetic specialists should consider these findings when counseling patients on periorbital rejuvenation.
An increase in MRD2 was observed in AI-enhanced photographs of both male and female participants. This aligns with prior research, which shows that desired photograph modifications often lead to an increased MRD2.3 Aesthetic trends today increasingly favor larger, more prominent eyes, creating a youthful and innocent look. Increasing MRD2 enhances the expressive, dramatic appearance of the eyes, a look popular in digital and social media. This contrasts with traditional lower eyelid surgeries, which typically reduce MRD2 by removing excess skin and tightening the lower lid for a smoother, more youthful appearance.3 The difference reflects evolving aesthetic preferences, especially in online spaces. An innovative option for achieving an increased MRD2 may be to consider targeted chemodenervation of the inferior orbicularis fibers, which could result in a wider palpebral fissure when smiling, without the risk of lid retraction at rest.5
Although these results provide valuable insights, this study has several limitations. There were inherent differences in the selfies submitted by participants because participants submitted selfie photographs in environments that varied in lighting and camera angle. In addition, this study population was predominantly younger and individuals with darker skin tones were not well represented, which may affect the applicability of the authors’ findings. Another limitation is that results of this study are specific to the Remini AI Photo Enhancer application; other AI systems may employ different algorithms and aesthetic priorities, leading to variations in the modifications they generate. It is important to note that the “Casual Headshot” pack was used to enhance all images, and a standardized model (one for male and one for female participants) served as the reference for the modifications. This feature is designed to create professional, polished headshots, which likely influenced the observed changes, such as the reduction in smile angle and dental show, aligning with the aesthetic goals of subtlety and balance typical of headshots. These findings might differ if other enhancement modes or applications were used, as they may prioritize different aesthetic goals. Another limitation is the use of Emotrics software to assess changes in facial measurements. As such, it is not capable of evaluating other factors, such as skin texture, overall facial proportions, or broader asymmetry changes, which were also altered by the AI platform. Because of these constraints, this study was limited in capturing these additional changes. Future studies using more comprehensive analysis tools that can assess these broader facial features would provide a more complete understanding of the modifications made by AI-enhanced images.
Future studies should replicate this research using digital camera portraits instead of smartphone selfies to determine if the changes in facial measurements are comparable. It is important to note that selfies themselves often distort facial proportions because of factors such as focal length, which can exaggerate central facial features. The Remini AI Photo Enhancer, as used in this study, directs users to upload selfies, and the authors aimed to replicate how the general public typically interacts with such AI platforms. As a result, findings of this study may be influenced by these selfie distortions, and future research could explore how AI-generated enhancements correct for these distortions, especially in comparison with more standardized images taken with DSLR cameras at a typical focal length. Moreover, further research is needed to evaluate how AI-enhanced images are perceived in attractiveness, both by human observers and AI software. Understanding how these images are rated by different entities can reveal discrepancies between human aesthetic preferences and AI’s interpretive capabilities. This insight will not only help clarify how well AI-generated enhancements align with human beauty standards but also inform future improvements in AI-driven photograph editing technologies.
Conclusion
The results of this analysis demonstrate that AI-enhanced selfies lead to significant changes in facial measurements. Notably, AI reduced brow height, commissure excursion, smile angle, and dental show, while increasing MRD2. These alterations were observed across both genders, with significant reductions in brow height specifically for male participants. These findings suggest that AI modifications prioritize features associated with more subtle, symmetrical expressions, aligning with current beauty trends. This focus on balanced and refined aesthetics underscores the growing influence of AI on modern beauty standards. By understanding these trends, surgeons can better tailor their procedures to align with evolving preferences shaped by social media, while also offering more informed consultations to help patients set realistic expectations for their treatments.
Read the article as originally posted on PubMed.