Beyond wheelchairs and blindfolds: Investigating disability stereotypes in T2I models with INCLUDE-BENCH

2026-07-09

AI Summary

This work addresses the absence of sociologically grounded evaluations of disability bias in current text-to-image (T2I) models. It introduces INCLUDE-BENCH, the first large-scale benchmark for assessing disability-related biases, comprising 119K images, multidimensional prompts, and both static and dynamic scenes. For the first time in this domain, the Stereotype Content Model (SCM) from social psychology is integrated to formulate an SCM Score for quantitative bias measurement. The study reveals that all 17 mainstream T2I models exhibit a strong overreliance on wheelchair imagery when generating disability-related content, resulting in markedly reduced representational diversity. Notably, higher text-image alignment exacerbates stereotypical portrayals, underscoring a critical tendency of these models to reproduce real-world disability biases.

Abstract

Text-to-image (T2I) models have been shown to exhibit social biases. Prior work has mainly focused on gender, skin tone, and cultural representation within restricted occupational associations, and emerging benchmarks increasingly incorporate these dimensions. However, disability remains systematically underexplored. Current evaluation practices often fail to align with sociologically grounded definitions of stereotyping, limiting principled assessment of representational harms toward people with disabilities (PWD). To address this, we introduce INCLUDE-BENCH, the first large-scale benchmark for evaluating disability-related bias in T2I models. INCLUDE-BENCH comprises 119K generated images based on prompt design across multiple bias dimensions and both static and dynamic contexts. We evaluate 15 open-source and 2 closed-source models. Our key findings reveal that: (1) mobility-impaired and default disability prompts predominantly yield wheelchair depictions across all models; (2) disability-conditioned generations consistently exhibit less diversity; (3) stereotypical portrayals demonstrate stronger disability-text alignment; and (4) we introduce the Stereotype Content Model (SCM) Score, demonstrating that T2I models reflect real-world stereotypical associations.

Problem

• disability
• stereotypes
• text-to-image models
• representational harms
• bias evaluation

Innovation

• INCLUDE-BENCH
• disability bias
• text-to-image models
• Stereotype Content Model
• representational harm