In the winter of 2022–23 I spent several months working with DALL·E 2, one of the first image generators open to the public. It is hard now to recall how startling it was: type a sentence and, seconds later, receive a photograph no one had taken. I turned it on subjects that run through my own work: olive groves and demolished houses in the West Bank, roadside shrines, a Taiwanese convenience store, the steel frames of Chinese-built monuments in Africa, German allotment gardens. I generated some 12,000 images, adjusting prompts the way a photographer adjusts light and framing, and kept very few.

What drew me was how photographic these early images could be. They were rough and unpredictable, closer to a found snapshot than to an illustration, and now and then something appeared that I had not asked for, a ghost in the machine. Assembled from the internet’s vast reservoir of pictures, they could feel less invented than remembered. The one thing I avoided was people: I didn’t like how the model rendered them, so these images are of places and things.

Within a year the tools had become far more controllable and polished. The strangeness did not disappear entirely, but it became a style to be chosen rather than something the machine produced on its own. After-Images is a record of that brief interval: seven typologies made with a technology at the moment it could still surprise.