Abstract
Generative AI (gAI) image models are increasingly being used as tools for conceptualisation and visualisation in design. These models are, however, not neutral owing to training datasets that prioritise Western aesthetics. Consequently, an emergent challenge is that AI-generated images tend to disregard nuanced representations. Li et al. (2024) echo this critique, acknowledging AI’s inability to comprehend cultural complexity, a critical issue in South Africa’s given its colonial and apartheid past. Accordingly, we use a relational design lens to critically examine how Stable Diffusion fails to engage with South Africa’s diverse cultural makeup; we present a qualitative content analysis of AI-generated datasets in response to communication design prompts. This paper contributes to discussions on AI’s role in shaping contemporary visual culture and its broader implications for design. Additionally, we explore how designers can influence AI training to foster more culturally inclusive and representative algorithms.
Keywords
Generative AI; Stable Diffusion; South African design; Representation; Colonial bias; Cultural narratives
DOI
https://doi.org/10.21606/nordes.2025.30
Citation
Rath, K., Cassim, F.,and Haese, A.(2025) Lost in generation: illustrating Stable Diffusion AI’s colonial bias in representing South African communication design, in Brandt, E., Markussen, T., Berglund, E., Julier, G., Linde, P. (eds.), Nordes 2025: Relational Design, 6-8 August, Oslo, Norway. https://doi.org/10.21606/nordes.2025.30
Creative Commons License
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License
Conference Track
Full Papers
Included in
Lost in generation: illustrating Stable Diffusion AI’s colonial bias in representing South African communication design
Generative AI (gAI) image models are increasingly being used as tools for conceptualisation and visualisation in design. These models are, however, not neutral owing to training datasets that prioritise Western aesthetics. Consequently, an emergent challenge is that AI-generated images tend to disregard nuanced representations. Li et al. (2024) echo this critique, acknowledging AI’s inability to comprehend cultural complexity, a critical issue in South Africa’s given its colonial and apartheid past. Accordingly, we use a relational design lens to critically examine how Stable Diffusion fails to engage with South Africa’s diverse cultural makeup; we present a qualitative content analysis of AI-generated datasets in response to communication design prompts. This paper contributes to discussions on AI’s role in shaping contemporary visual culture and its broader implications for design. Additionally, we explore how designers can influence AI training to foster more culturally inclusive and representative algorithms.