A city that never existed, and is unmistakably London.

We curated two archives of the city, historic and modern, and trained a generative adversarial network (DCGAN) on them in raw Python. No interface, no presets. It returned streets no one has ever photographed.

Months later, Photoshop shipped its first GAN-powered filter and pointed it at faces. Same class of model, entirely different subject: evidence of how far the method reaches. We had already pointed it at the city.

If a place can be learned, it can be authored.