Textual and Visual Interpretation in a Text-to-Image Accelerated Architectural Design Process

The introduction of text-to-image diffusion models poses the question whether design processes could be accelerated: Can rapid intuitive image generation replace traditional representational methods? This paper studies the results of an accelerated design process, conducted by 68 master students of architecture divided into groups of 4. The students are presented with a design brief and asked to use Midjourney during the sketch design phase. The produced images would represent their architectural translation of the brief that should lead to a realistic ‘buildable’ project. This paper studies the design decisions and design criteria through the collected text prompts, images and questions into the used vocabulary, experiences and decision-making process, leading to the definition of textual and visual interpretation moments. Furthermore, the relationship between the textual and visual interpretation moments and their respective challenges are identified as a basis for further research into sketch design support and automation with diffusion models. Our paper provides insights into the potential of accelerated workflows in architectural design, identifying key inflection points in the process. This is significant as it provides part of the basis for understanding the role diffusion models might have in a 4^th industrial revolution workflow of architectural design.