Thursday 20 March 2025
Scientists have been working on creating synthetic portraits that accurately depict human faces at specific ages, but a recent study suggests that this technology still has a long way to go.
The researchers used three state-of-the-art text-to-image generation models to create images of individuals with different nationalities, ages, and genders. They then evaluated these images using two established age-estimation systems to see how accurately they could predict the age of each person.
The results were mixed. While the models were able to consistently generate faces that reflected different identities, they struggled to capture subtle changes in facial appearance as people aged. The study found that the accuracy of the age estimates decreased significantly for older individuals, with errors ranging from a few years to over a decade.
One of the challenges facing this technology is the difficulty of accurately rendering facial features that change dramatically with age, such as wrinkles and age-related hair loss. The models also tended to produce images that were more youthful-looking than their real-life counterparts, which can lead to inaccuracies in age estimation.
Another issue is bias. The researchers found that certain demographic groups, such as older adults and women, were more likely to be misestimated than others. This highlights the need for greater diversity and representation in the training data used to develop these models.
Despite these limitations, the study suggests that text-to-image generation technology has potential applications in areas such as identity verification, data augmentation, and exploratory analysis. However, it is crucial to address the biases and inaccuracies identified in this research before this technology can be widely adopted.
To improve the accuracy of synthetic portraits, researchers may need to develop more sophisticated models that better capture the subtleties of human facial aging. They may also need to incorporate additional data sources, such as real-world images of individuals at different ages, to reduce bias and increase diversity in their training sets.
Ultimately, the development of realistic synthetic portraits is a complex task that requires careful consideration of both technical and social factors. As researchers continue to explore this technology, it will be important to prioritize accuracy, fairness, and transparency to ensure that these models are used responsibly and benefit society as a whole.
Cite this article: “Limitations of Synthetic Portrait Technology”, The Science Archive, 2025.
Synthetic Portraits, Age Estimation, Facial Recognition, Text-To-Image Generation, Deep Learning Models, Accuracy, Bias, Identity Verification, Data Augmentation, Exploratory Analysis
