Generative Architectural Ideation
Speculative Architectural Ideation using generative adversarial network (GAN).
2021 / Workshop
Machine Learning
Mosse Sjaastad
Lars Marcus Vedeler

A one-week workshop exploring Machine Learning and Computer Vision through small experiments.

"Speculative ideation" in Architecture encourages exploration through different mediums. The process intends to invigorate foundational skills and knowledge. It includes exercises drawn upon an analytical approach to materiality, composition, expression and form.

During this workshop I explored the possibility for a Generative Adversarial Network, trained on a case specific dataset, to contribute to this process in any meaningful way. The result shows that the model is capable of generating some interesting sketches on form, composition and expression.

To me, this workshop ended up being evidence on how AI technologies might be used as a creative tool. Architects could draw insight and inspiration from the sometimes-unexpected outcome of a machine.

The collected dataset for this workshop consisted of 217 unique images of residential house facades.
The StyleGAN-model was then trained on this set of images for 2500 steps, using the product RunwayML.
The result is a ML-model that generates unique and unconventional structures resembling that of house facades. The architect could use the generated images to challenge existing ideas and invigorate the creative process.