DIA GeD

Welcome to the DIA GeD course, part of the CAD/Logic didactic offer. This course introduces Generative Design for Architecture, the same frontier of the digital revolution that is about to disrupt architecture as a process as well as a product. The architect is perhaps the only artist who produces art without manipulating physical matter. The immateriality of architectural design has been reflected in the use of computers since the end of the twentieth century, when Computer-Aided-Design (first) and Building Information Modeling (later) even replaced paper as the physical support for information exchange. In May 2023, NVidia's CEO likened the emergence of ChatGPT to the groundbreaking iPhone moment of 2007, heralding a significant shift akin to the smartphone revolution. Alongside Large Language Models (LLMs) like ChatGPT, Text-to-Image generators represent another significant application of Artificial Generative Intelligence (AGI), increasingly integrated into educational settings, particularly sought after by students.

Generative design is an exploration process based on the power of computers. It transforms the design journey into the construction and exploration of a design space made up of hundreds or thousands of computer-generated options that solve the same problem. Already established in higher degree of freedom design (i.e., product design or manufacturing), this iterative design process pushes the limits of dimensionality and paves the way for data-driven collaborative decision-making in the practice of architecture. Implementing generative AI in architecture means to strike a deal between control over the model and freedom for the GAI to explore and stimulate the unconscious dimension of the design process.

The course presents an infrastructure of freeware tools to construct and explore design spaces with Grasshopper for Rhino to be supplemented by AGI using Stable Diffusion. This open-source deep learning Text-to-Image model launched in 2022 offers precise control over the diffusion process running locally through ComfyUI, a user-friendly graphical interface that uses nodes and graphs instead of line-coding, which can read the Grasshopper input. Reuse of data is the new paradigm for digital design processes: to lead by example, DIAGeD will inherit data from the previous edition and expand in the summer term 2024 on the scale of architectural skin elements (facade, roof, etc.) in the context of an intervention on Dessau's built heritage, in particular the extension of the zoo on the ex-Betriebsbanhof site, with the aim of trade-off between different design dimensions such as sustainability, statics, aesthetics and so on. For the first time, DIAGeD will propose a single team design approach using Rhino's worksession functionality and Rhino.Inside®.Revit is introduced to streamline Generative Design results from and into BIM, while Generative AI with Stable Diffusion  and 3D printing of small mockups in the robotics lab will pave the way from digital to analog that characterize the course production. 

DIAGeD will propose hands-on challenges in class, while assignments will be distributed afterwards and finally presented the following week. Points will be awarded for both submissions and presentations. Submission of all assignments is sufficient to pass the course with a 4.00. Presentations and performances reward a grade up to a 2.00. To earn the highest grade, a final submission is required, which includes the delivery of a design space as taught in class.