Prompting Away Stereotypes? Evaluating Bias in Text-to-Image Models for Occupations

2025/08/31 by Shaina Raza, Raza, Shaina, Powers, Maximus +5
Computer Science · Social Sciences · #Baseline (sea) #Benchmark (surveying) #Code (set theory) #Computation and Language (cs.CL) #Computational and Text Analysis Methods #Ethics and Social Impacts of AI #FOS: Computer and information sciences #Hate Speech and Cyberbullying Detection #Race (biology) #Salient #Software #Task (project management) #Transparency (behavior)

paper · pdf · doi:10.48550/arxiv.2509.00849

published in arXiv (Cornell University) (Cornell University)

openalex publication_date 2025/08/31 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/05

Abstract

Text-to-Image (TTI) models are powerful creative tools but risk amplifying harmful social biases. We frame representational societal bias assessment as an image curation and evaluation task and introduce a pilot benchmark of occupational portrayals spanning five socially salient roles (CEO, Nurse, Software Engineer, Teacher, Athlete). Using five state-of-the-art models: closed-source (DALLE 3, Gemini Imagen 4.0) and open-source (FLUX.1-dev, Stable Diffusion XL Turbo, Grok-2 Image), we compare neutral baseline prompts against fairness-aware controlled prompts designed to encourage demographic diversity. All outputs are annotated for gender (male, female) and race (Asian, Black, White), enabling structured distributional analysis. Results show that prompting can substantially shift demographic representations, but with highly model-specific effects: some systems diversify effectively, others overcorrect into unrealistic uniformity, and some show little responsiveness. These findings highlight both the promise and the limitations of prompting as a fairness intervention, underscoring the need for complementary model-level strategies. We release all code and data for transparency and reproducibility https://github.com/maximus-powers/img-gen-bias-analysis.

Citations

• Beyond Content: How Grammatical Gender Shapes Visual Representation in Text-to-Image Models
• Evaluating Fairness in Large Vision-Language Models Across Diverse Demographic Attributes and Prompts
• Quantifying Bias in Text-to-Image Generative Models
• TIBET: Identifying and Evaluating Biases in Text-to-Image Generative Models
• Fair Text-to-Image Diffusion via Fair Mapping
• Holistic Evaluation of Text-To-Image Models
• The Bias Amplification Paradox in Text-to-Image Generation
• Stable Bias: Analyzing Societal Representations in Diffusion Models
• Aligning Text-to-Image Models using Human Feedback
• LAION-5B: An open large-scale dataset for training next generation image-text models
• Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback
• DALL-Eval: Probing the Reasoning Skills and Social Biases of Text-to-Image Generation Models
• Assessing Social and Intersectional Biases in Contextualized Word Representations
• FairT2I: Mitigating Social Bias in Text-to-Image Generation via Large Language Model-Assisted Detection and Attribute Rebalancing
• Who is Responsible? The Data, Models, Users or Regulations? A Comprehensive Survey on Responsible Generative AI for a Sustainable Future
• Role congruity theory of prejudice toward female leaders.

Related

• Hidden Bias in the Machine: Stereotypes in Text-to-Image Models
• Stable Bias: Analyzing Societal Representations in Diffusion Models
• T2IBias: Uncovering Societal Bias Encoded in the Latent Space of Text-to-Image Generative Models
• New Job, New Gender? Measuring the Social Bias in Image Generation Models
• Automated Evaluation of Gender Bias Across 13 Large Multimodal Models