Home // AIMEDIA 2025, The First International Conference on AI-based Media Innovation // View article

Evaluating Diffusion-Based Image Generation for Easy Language Accessibility

Authors:
Christoph Johannes Weber
Dominik Beyer
Sylvia Rothe

Keywords: image generation; accessibility; easy language.

Abstract:
Easy Language is a linguistic resource designed to facilitate comprehension for individuals with learning impairments and non-native speakers. Utilizing simplification of text, the restriction of vocabulary, and layout adjustments, Easy Language texts are constructed to ensure accessibility. Furthermore, Easy Language texts have the capacity to incorporate visual aids, such as images or symbols, to enhance comprehension. Although the use of imagery has been shown to improve understanding, it remains unclear whether visuals generated by artificial intelligence (AI) can meet the specific stylistic and semantic requirements of Easy Language. This paper investigates the potential of diffusion-based image generation to address these needs. A Stable Diffusion model was fine-tuned to produce images in a minimal, symbol-like style. Two user studies were conducted to assess the model’s ability to replicate a consistent visual style and to determine whether the generated images effectively conveyed the intended meanings. The results show that participants were generally unable to distinguish AI-generated from original symbols and correctly interpreted most of the illustrated concepts. The findings suggest that diffusion models, when properly fine-tuned, are capable of producing illustrations that align with the stylistic conventions and semantic clarity required in Easy Language. However, certain abstract or emotionally nuanced concepts remain challenging to represent accurately. These results indicate that, when guided by stylistic constraints, AI-generated visuals can offer a scalable approach to producing accessible visual content.

Pages: 32 to 40

Copyright: Copyright (c) IARIA, 2025

Publication date: July 6, 2025

Published in: conference

ISBN: 978-1-68558-330-9

Location: Venice, Italy

Dates: from July 6, 2025 to July 10, 2025