AI IMAGE GENERATION AND ITS POSSIBLE CONTRIBUTIONS IN ARCHITECTURAL LANGUAGE

AI, and specifically Machine Learning methodologies, can help the architect's imagination in the design process or the urban planner to develop ideas about urban planning. This paper focuses on generative ML methodologies that generate images from input datasets. The machine is trained to create patterns that help generate new images. In all of these methodologies, the Generative Adversarial Networks (GANs) model is used, with specific details for each case. The idea is to highlight their specifics, both in concept and in implementation, and to test different metrics for assessing the accuracy of the generative process. The input datasets are facade images, their sketches, or the combinations of both, and as a result, new images can be generated. The machine learning techniques are used to help us interpret architectural historical concepts, such as the relationship between the natural and the customary. Traditionally, the natural is represented by geometry, while the customary has stood for inherited stylistic languages.