Implementing a Generative AI Model in the Architectural Concept Development: Context-based Research

Abstract

In the context of the continual evolution of advanced architectural design tools, driven by computational and artificial intelligence technologies, the practice of design can now benefit from contextually-driven and iterative design workflows. Thus, computational and AI tools have huge potential to enhance the conceptual design processes in architecture, due to their advanced workflow qualities. Node-based interfaces of these tools serve as a catalyst, enabling workflows that are adaptable to different requirements and needs of the design processes.This paper explores juxtaposing node-based computational design and AI tools in a dependent workflow process using Grasshopper3D and ComfyUI, within the framework of an experimental architectural concept development process,through a qualitative Research-through-Design methodology (RtD). Through a series of controlled experiments, multiple architectural design schemes are generated and categorized into four distinct visual identities: blob, biophilic, vernacular, and modern. These categories serve as a comparative framework to evaluate the adaptability and responsiveness of the integrated workflow across different design intentions. The process emphasizes iteration and variation, but with contextual responsiveness, testing a more streamlined refinement process of design schemes.The findings demonstrate that integrating node-based computational scripting with AI-driven workflows enhances the capacity for rapid iteration, supports the generation of diverse and contextually relevant architectural concepts, and enables more fluid transitions between abstract and representational stages of design. This research contributes to the advancement of architectural design methodologies by implementing a hybrid workflow model that bridges computational logic and generative AI capabilities, showcasing new potentials for concept development in contemporary architectural practice. Node-based interfaces of these tools act as a catalyst to allow more custom and interconnected workflows. Moreover, it is important to test new design potentials that these tools contribute to when used in parallel for the conceptual design process. This thesis explores juxtaposing node-based computational design and AI tools in a dependent workflow process using Grasshopper3D and ComfyUI, within the framework of an experimental architectural concept development process, through a qualitative Research-through-Design methodology(RtD). By experimenting with their workflows, the aim is to formulate different design schemes that are contextually relevant but also achieved iteratively. Visualizing abstract forms to test architectural forms relevant to four styles, which are blob, biophilic, vernacular, and modern architectural styles. Gathering the data about these outputs and analyzing the overall experiment. Ultimately, this thesis achieves iterative, data-informed design schemes with detailed architectural features. Aiding contextually relevant but also architecturally detailed design schemes during the conceptual phase of the architectural design process.