Akademik Veri Yönetim Sistemi
V. International Architectural Sciences Congress (IArcSC-2025), Antalya, Türkiye, 29 - 31 Ekim 2025, ss.91, (Özet Bildiri)
This study aims to examine the contributions of generative artificial intelligence (GAI) models to the processes of
architectural form generation. In particular, the models DALL·E (OpenAI), Midjourney, and Stable Diffusion,
which generate images from text prompts, are comparatively analyzed. The primary objective is to evaluate the
potential of these models in terms of conceptual design, formal diversity, visual quality, and user control, and to
determine their level of integration into the architectural design process.
A qualitative comparative methodology is adopted in this research. Architectural design prompts with identical
or similar content were used with each model to generate visual outputs. These outputs were assessed based on
criteria such as formal creativity, technical precision, aesthetic quality, and ease of use. Additionally, functional
aspects including accessibility, controllability, and applicability in educational and research environments were
analyzed.
The findings indicate that DALL·E provides a balanced approach between realistic and conceptual visualization;
Midjourney stands out in producing aesthetically rich, abstract, and creative forms; while Stable Diffusion
emerges as a strong alternative for experimental and research-oriented architectural productions, owing to its
open-source structure and customizability. Moreover, Stable Diffusion demonstrates higher levels of prompt
sensitivity and responsiveness to technical detail.
In conclusion, generative AI models can be integrated as creative and supportive tools into the architectural formgeneration
process. However, the choice of model should vary depending on the design objective and contextual
needs.
Keywords: Artificial intelligence, architectural form generation, generative ai models, DALL·E, midjourney,
stable diffusion, visualization in the design process.