MineTheGap: Automatic Mining of Biases in Text-to-Image Models

2025-12-15

AI Summary

Text-to-image (TTI) models often exhibit latent societal biases—such as racial stereotypes in occupational prompts—and generate redundant outputs when processing semantically ambiguous prompts, thereby compromising fairness and user experience. To address this, we propose the first proactive bias mining framework tailored for ambiguous prompts. Our method introduces a multimodal distribution alignment–based bias scoring function to quantify inconsistency between LLM-generated prompt variants and their corresponding image distributions. Further, we design a genetic algorithm–driven iterative prompt optimization mechanism that actively discovers bias-triggering prompts rather than passively detecting them. Evaluated on standard bias benchmark datasets, our approach significantly improves recall of biased prompts. We publicly release our code and curated case studies to ensure reproducible evaluation.

Technology Category

• Computer Vision: Bias, Fairness & Privacy
• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy
• Machine Learning: Ethics, Bias, and Fairness

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Social Networks and Social Media: Fairness and bias in social network and social media analysis
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information

Abstract

Text-to-Image (TTI) models generate images based on text prompts, which often leave certain aspects of the desired image ambiguous. When faced with these ambiguities, TTI models have been shown to exhibit biases in their interpretations. These biases can have societal impacts, e.g., when showing only a certain race for a stated occupation. They can also affect user experience when creating redundancy within a set of generated images instead of spanning diverse possibilities. Here, we introduce MineTheGap - a method for automatically mining prompts that cause a TTI model to generate biased outputs. Our method goes beyond merely detecting bias for a given prompt. Rather, it leverages a genetic algorithm to iteratively refine a pool of prompts, seeking for those that expose biases. This optimization process is driven by a novel bias score, which ranks biases according to their severity, as we validate on a dataset with known biases. For a given prompt, this score is obtained by comparing the distribution of generated images to the distribution of LLM-generated texts that constitute variations on the prompt. Code and examples are available on the project's webpage.

Problem

• Automatically mines prompts causing biases in text-to-image models
• Addresses societal impacts like racial stereotyping in generated images
• Reduces redundancy by identifying prompts that limit output diversity

Innovation

• Genetic algorithm refines prompts to expose biases
• Novel bias score ranks severity by comparing image distributions
• Automatically mines prompts causing biased TTI model outputs