Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Can Text-to-Image Generative Models Accurately Depict Age? A Comparative Study on Synthetic Portrait Generation and Age Estimation

Alexey A. Novikov, Miroslav Vranka, François David, Artem Voronin

TL;DR

This study systematically evaluates whether text-to-image generative models can accurately depict age in synthetic portraits across $212$ nationalities, $30$ ages, and balanced genders, using two age estimators as ground truth. It compares Flux, SD3.5, and Epic Realism, finding that while identity cues are generally preserved, precise age depiction is inconsistent and biased, with Epic Realism showing the strongest age alignment but notable errors at older ages. The work highlights significant biases and outliers in synthetic data, cautions against using such images for high-stakes age tasks without filtering, and discusses mitigations through targeted training, prompt engineering, and post-hoc correction. It also addresses ethical, privacy, and security considerations for deploying synthetic age data in practical biometric workflows.

Abstract

Text-to-image generative models have shown remarkable progress in producing diverse and photorealistic outputs. In this paper, we present a comprehensive analysis of their effectiveness in creating synthetic portraits that accurately represent various demographic attributes, with a special focus on age, nationality, and gender. Our evaluation employs prompts specifying detailed profiles (e.g., Photorealistic selfie photo of a 32-year-old Canadian male), covering a broad spectrum of 212 nationalities, 30 distinct ages from 10 to 78, and balanced gender representation. We compare the generated images against ground truth age estimates from two established age estimation models to assess how faithfully age is depicted. Our findings reveal that although text-to-image models can consistently generate faces reflecting different identities, the accuracy with which they capture specific ages and do so across diverse demographic backgrounds remains highly variable. These results suggest that current synthetic data may be insufficiently reliable for high-stakes age-related tasks requiring robust precision, unless practitioners are prepared to invest in significant filtering and curation. Nevertheless, they may still be useful in less sensitive or exploratory applications, where absolute age precision is not critical.

Can Text-to-Image Generative Models Accurately Depict Age? A Comparative Study on Synthetic Portrait Generation and Age Estimation

TL;DR

This study systematically evaluates whether text-to-image generative models can accurately depict age in synthetic portraits across nationalities, ages, and balanced genders, using two age estimators as ground truth. It compares Flux, SD3.5, and Epic Realism, finding that while identity cues are generally preserved, precise age depiction is inconsistent and biased, with Epic Realism showing the strongest age alignment but notable errors at older ages. The work highlights significant biases and outliers in synthetic data, cautions against using such images for high-stakes age tasks without filtering, and discusses mitigations through targeted training, prompt engineering, and post-hoc correction. It also addresses ethical, privacy, and security considerations for deploying synthetic age data in practical biometric workflows.

Abstract

Text-to-image generative models have shown remarkable progress in producing diverse and photorealistic outputs. In this paper, we present a comprehensive analysis of their effectiveness in creating synthetic portraits that accurately represent various demographic attributes, with a special focus on age, nationality, and gender. Our evaluation employs prompts specifying detailed profiles (e.g., Photorealistic selfie photo of a 32-year-old Canadian male), covering a broad spectrum of 212 nationalities, 30 distinct ages from 10 to 78, and balanced gender representation. We compare the generated images against ground truth age estimates from two established age estimation models to assess how faithfully age is depicted. Our findings reveal that although text-to-image models can consistently generate faces reflecting different identities, the accuracy with which they capture specific ages and do so across diverse demographic backgrounds remains highly variable. These results suggest that current synthetic data may be insufficiently reliable for high-stakes age-related tasks requiring robust precision, unless practitioners are prepared to invest in significant filtering and curation. Nevertheless, they may still be useful in less sensitive or exploratory applications, where absolute age precision is not critical.

Paper Structure

This paper contains 24 sections, 2 figures, 4 tables.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORK
  3. METHODOLOGY
  4. Dataset and Prompt Collection
  5. Prompt Design
  6. Generative Models
  7. Flux
  8. SD35
  9. EpicRealism
  10. Age Estimation
  11. RESULTS AND DISCUSSION
  12. Comparison of Prompt Age vs. Estimated Age
  13. Analysis by Demographic Subgroups
  14. JAM + Flux
  15. JAM + SD 3.5
  16. ...and 9 more sections

Figures (2)

  • Figure 1: Examples of generated images using different methods: Flux (a, b), Stable Diffusion 3.5 Large (c, d), and SDXL Epic Realism (e, f).
  • Figure 2: Prompt vs. Predicted Age plots with regression lines for two T2I settings, Flux (left) and Epic Realism (right). Each point represents a single synthetic face, and the diagonal line indicates perfect age alignment.