Playing Bias: Epistemic Games for Interpreting Text-to-Image Models

Text-to-image (T2I) models often produce stereotypical images when stereotype-relevant attributes are left unspecified. We ask whether explicit requests can overcome these defaults, especially when the requested attribute contradicts the stereotype. We introduce \mbox{IMPLICIT-Bench}, a challenge benchmark of 1,831 prompt triplets built on a bias knowledge graph (KG) of stereotypes recorded in StereoSet and CrowS-Pairs. Each triplet describes one scene with a neutral prompt, a stereotype prompt, and an anti-stereotype prompt. The neutral prompt measures the model's default, and the two explicit prompts measure whether the model can follow a request on either side of the stereotype. On this benchmark, Qwen-Image depicts the stereotype in 52% of neutral images and the anti-stereotype in 5%. On both Qwen-Image and Stable Diffusion~3, explicit requests that contradict this default succeed far less often than requests that agree with it. Where Qwen-Image's default is strongest, explicitly requesting the stereotype succeeds in 97% of cases, whereas requesting the anti-stereotype succeeds in only 54%. In these scenes, 31% of the images for anti-stereotype requests show the stereotypical attribute instead. We further test whether prompt rewriting and embedding steering can reduce the default on neutral prompts. With the prompt's own KG unit, the best configurations improve a combined bias and alignment score by up to 9.5 points, and with a unit retrieved from the bias KG, by at most 2.9. Debiasing thus depends on knowing which stereotype a prompt evokes.

Text-to-image (T2I) models, and their encoded biases, increasingly shape the visual media the public encounters. While researchers have produced a rich body of work on bias measurement, auditing, and mitigation in T2I systems, those methods largely target technical stakeholders, leaving a gap in public legibility. We introduce GLEaN (Generative Likeness Evaluation at N-Scale), a portrait-based explainability pipeline designed to make T2I model biases visually understandable to a broad audience. GLEaN comprises three stages: automated large-scale image generation from identity prompts, facial landmark-based filtering and spatial alignment, and median-pixel composition that distills a model's central tendency into a single representative portrait. The resulting composites require no statistical background to interpret; a viewer can see, at a glance, who a model 'imagines' when prompted with 'a doctor' versus a 'felon.' We demonstrate GLEaN on Stable Diffusion XL across 40 social and occupational identity prompts, producing composites that reproduce documented biases and surface new associations between skin tone and predicted emotion. We find in a between-subjects user study (N = 291) that GLEaN portraits communicate biases as effectively as conventional data tables, but require significantly less viewing time. Because the method relies solely on generated outputs, it can also be replicated on any black-box and closed-weight systems without access to model internals. GLEaN offers a scalable, model-agnostic approach to bias explainability, purpose-built for public comprehension, and is publicly available at https://github.com/cultureiolab/GLEaN.

Text-to-image (T2I) generative models achieve impressive visual fidelity but inherit and amplify demographic imbalances and cultural biases embedded in training data. We introduce T2I-BiasBench, a unified evaluation framework of thirteen complementary metrics that jointly captures demographic bias, element omission, and cultural collapse in diffusion models - the first framework to address all three dimensions simultaneously. We evaluate three open-source models - Stable Diffusion v1.5, BK-SDM Base, and Koala Lightning - against Gemini 2.5 Flash (RLHF-aligned) as a reference baseline. The benchmark comprises 1,574 generated images across five structured prompt categories. T2I-BiasBench integrates six established metrics with seven additional measures: four newly proposed (Composite Bias Score, Grounded Missing Rate, Implicit Element Missing Rate, Cultural Accuracy Ratio) and three adapted (Hallucination Score, Vendi Score, CLIP Proxy Score). Three key findings emerge: (1) Stable Diffusion v1.5 and BK-SDM exhibit bias amplification (>1.0) in beauty-related prompts; (2) contextual constraints such as surgical PPE substantially attenuate professional-role gender bias (Doctor CBS = 0.06 for SD v1.5); and (3) all models, including RLHF-aligned Gemini, collapse to a narrow set of cultural representations (CAS: 0.54-1.00), confirming that alignment techniques do not resolve cultural coverage gaps. T2I-BiasBench is publicly released to support standardized, fine-grained bias evaluation of generative models. The project page is available at: https://gyanendrachaubey.github.io/T2I-BiasBench/