Exposing Hidden Biases in Text-to-Image Models via Automated Prompt Search

2025-12-09

AI Summary

Text-to-image (TTI) diffusion models exhibit high visual fidelity but harbor implicit societal biases along sensitive attributes such as gender, race, and age. Existing debiasing approaches rely on manually curated or LLM-generated prompt sets, suffering from limited coverage and inability to uncover unforeseen bias-triggering prompts. This paper proposes Bias-Guided Prompt Search (BGPS), the first framework enabling interpretable, user-controllable, and high-coverage bias detection. BGPS leverages attribute classifiers to guide an LLM in searching for natural-language prompts—within a neutral prompt space—that maximally amplify bias. It jointly exploits internal TTI model representations and LLM-generated feedback, refining the search via decoding-level control. Evaluated on Stable Diffusion 1.5 and state-of-the-art debiased models, BGPS successfully uncovers diverse latent biases, significantly degrading fairness metrics while demonstrating superior prompt interpretability and search efficacy.

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

Abstract

Text-to-image (TTI) diffusion models have achieved remarkable visual quality, yet they have been repeatedly shown to exhibit social biases across sensitive attributes such as gender, race and age. To mitigate these biases, existing approaches frequently depend on curated prompt datasets - either manually constructed or generated with large language models (LLMs) - as part of their training and/or evaluation procedures. Beside the curation cost, this also risks overlooking unanticipated, less obvious prompts that trigger biased generation, even in models that have undergone debiasing. In this work, we introduce Bias-Guided Prompt Search (BGPS), a framework that automatically generates prompts that aim to maximize the presence of biases in the resulting images. BGPS comprises two components: (1) an LLM instructed to produce attribute-neutral prompts and (2) attribute classifiers acting on the TTI's internal representations that steer the decoding process of the LLM toward regions of the prompt space that amplify the image attributes of interest. We conduct extensive experiments on Stable Diffusion 1.5 and a state-of-the-art debiased model and discover an array of subtle and previously undocumented biases that severely deteriorate fairness metrics. Crucially, the discovered prompts are interpretable, i.e they may be entered by a typical user, quantitatively improving the perplexity metric compared to a prominent hard prompt optimization counterpart. Our findings uncover TTI vulnerabilities, while BGPS expands the bias search space and can act as a new evaluation tool for bias mitigation.

Problem

• Automatically generates prompts to expose hidden biases in text-to-image models
• Discovers subtle, undocumented biases that degrade fairness in diffusion models
• Expands bias search space as an evaluation tool for bias mitigation

Innovation

• Automated prompt search to expose biases
• Attribute classifiers steer LLM decoding
• Generates interpretable prompts improving bias detection