Experiment and Evaluation of Architectural Image Generation through Artificial Intelligence-Based Text Image Generation Tool
Dong Ho Lee, Sung Hak Ko · KIEAE Journal · 2023
Purpose: The accelerating speed of technological advancement means that AI technology cannot be isolated from our lives. And artists have already started to exchange a lot of information about using AI image generators. By comparison, there is a lack of research on AI image generators for architecture. Therefore, to highlight the possibility of an AI image generator based on a text-to-image model for application and utilization in actual architecture, the paper proceeded with the process of testing and evaluating. Method: Based on previous research, the paper outlined a new algorithmic methodology that is divided into a concept extraction phase and an architecturalization phase to generate creative concepts and architectural images. Each process of methodology was subjected to auxiliary experiments to ensure it worked effectively, and numerous image generation experiments were conducted to analyze and generate the result of methodology. Result: It was possible to derive and evaluate meaningful analysis results that can be used for other research and new methodology. And through a feedback loop process that transforms and repeats the entire process, it was possible to create images with creative colors and shapes that can be used in actual architecture. Finally, based on the limitations and possibilities of the current technology level, the paper predicted the future prospects of AI image generators in architecture and suggested the need for related follow-up studies.