The realm of architectural design is continually evolving, driven by advancing technology and desiring to push the boundaries of creativity. Among these technological advancements, image generators have emerged as a promising tool for designers, offering efficiency and inspiration in generating design ideas, providing potentially wide range of design concepts within limited timeframes. However, it is crucial to critically evaluate the quality of generative images, as this novel technology challenges traditional design methodologies and raises questions about the level of training and limitations within machine learning algorithms. Hence, this research project aims to experimentally investigate the translation process from 2D images to 3D models in architectural design, with the objective of identifying the gaps in this transition and uncovering new possibilities for constructing a novel typology of architectural elements.