Prompting Away Stereotypes? Evaluating Bias in Text-to-Image Models for Occupations
Not on the map. Why?
Abstract (arXiv)
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.
Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Builds on 17 of its 35 references that are in the map. First posted on arXiv 31 Aug 2025.
Builds on
17 of 35 references.
References · 17
- 2022LAION-5B: An open large-scale dataset for training next generation image-text modelscited by 4,131
- 2022Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedbackcited by 3,634
- 2023Aligning Text-to-Image Models using Human Feedbackcited by 418
- 2022DALL-Eval: Probing the Reasoning Skills and Social Biases of Text-to-Image Generation Modelscited by 230
- 2023Stable Bias: Analyzing Societal Representations in Diffusion Modelscited by 225
- 2020Social Biases in NLP Models as Barriers for Persons with Disabilitiescited by 182
- 2023Holistic Evaluation of Text-To-Image Modelscited by 160
- 2019Assessing Social and Intersectional Biases in Contextualized Word Representationscited by 134
- 2023The Bias Amplification Paradox in Text-to-Image Generationcited by 69
- 2023TIBET: Identifying and Evaluating Biases in Text-to-Image Generative Modelscited by 51
- 2022Gender and Racial Bias in Visual Question Answering Datasetscited by 39
- 2022Representation Bias in Data: A Survey on Identification and Resolution Techniquescited by 37
5 more
- 2023Fair Text-to-Image Diffusion via Fair Mappingcited by 31
- 2023Smiling Women Pitching Down: Auditing Representational and Presentational Gender Biases in Image Generative AIcited by 19
- 2023Quantifying Bias in Text-to-Image Generative Modelscited by 16
- 2024Evaluating Fairness in Large Vision-Language Models Across Diverse Demographic Attributes and Promptscited by 15
- 2025Beyond Content: How Grammatical Gender Shapes Visual Representation in Text-to-Image Modelscited by 4
Built on by
Citing papers per year: 2026: 1.
1 paper.
Similar, not cited
Close in wording, no citation either way
- 2023Social Biases through the Text-to-Image Generation Lens
- 2025Automated Evaluation of Gender Bias Across 13 Large Multimodal Models
- 2025Hidden Bias in the Machine: Stereotypes in Text-to-Image Models
- 2023Finetuning Text-to-Image Diffusion Models for Fairness
- 2023Stable Diffusion Exposed: Gender Bias from Prompt to Image
- 2026BAFIS: Dataset + Framework to assess occupational Bias and Human Preference in modern Text-to-image Models
- 2024New Job, New Gender? Measuring the Social Bias in Image Generation Models
- 2026Newer Is Not Fairer: Gender Stereotyping in Text-to-Image AI Across Model Generations