Who Defines Fairness? Target-Based Prompting for Demographic Representation in Generative Models

TL;DR

This paper introduces a lightweight, inference-time method for mitigating demographic bias in text-to-image models by adjusting prompts according to user-defined fairness specifications, without retraining the models.

Contribution

It presents a novel prompt-level intervention framework that allows users to specify fairness criteria and guides demographic representation in generated images.

Findings

The method effectively shifts skin-tone distributions towards target fairness specifications.

It reduces deviation from target demographic distributions across multiple prompts and contexts.

The approach enhances transparency and user control over fairness in generative AI outputs.

Abstract

Text-to-image(T2I) models like Stable Diffusion and DALL-E have made generative AI widely accessible, yet recent studies reveal that these systems often replicate societal biases, particularly in how they depict demographic groups across professions. Prompts such as 'doctor' or 'CEO' frequently yield lighter-skinned outputs, while lower-status roles like 'janitor' show more diversity, reinforcing stereotypes. Existing mitigation methods typically require retraining or curated datasets, making them inaccessible to most users. We propose a lightweight, inference-time framework that mitigates representational bias through prompt-level intervention without modifying the underlying model. Instead of assuming a single definition of fairness, our approach allows users to select among multiple fairness specifications-ranging from simple choices such as a uniform distribution to more complex definitions informed by a large language model(LLM) that cites sources and provides confidence estimates. These distributions guide the construction of demographic specific prompt variants in the corresponding proportions, and we evaluate alignment by auditing adherence to the declared target and measuring the resulting skin tone distribution rather than assuming uniformity as 'fairness'. Across 36 prompts spanning 30 occupations and 6 non-occupational contexts, our method shifts observed skin-tone outcomes in directions consistent with the declared target, and reduces deviation from targets when the target is defined directly in skin-tone space(fallback). This work demonstrates how fairness interventions can be made transparent, controllable, and usable at inference time, directly empowering users of generative AI.