GAN in the Design Office

AI machine learning technologies evolve in a fast pace. Generative Advasarial Network (GAN) images, which aren’t just visual representations of neural networks, are generative graphics. Using computer graphics and machine learnings, they can create real-looking visuals of non-existing objects. The invention was originally developed by Ian Goodfellow and later popularized by large tech companies.

The AI models aid architects in the communication of concepts under a reduced timespan, saving the effort to model landscape elements with traditional tools.

The science behind the machine learning still develops rapidly, resulting in significant improvements each year. In its current state, GAN technology is ready to become part of the architect’s workflow.

Context

The presented images are generated using NVIDIA’s GauGAN model, that turns simple stroke brushes into complex landscape visuals. Then the raw image is treated with simple colour correction. Single pre-photographed content is added to the collage to give the visual character, to let the GAN image stand out from its generated siblings.