How AI prompting turned writerly description into an everyday skill

To generate a satisfactory AI image, we must do for a machine what novelists have done for readers: translate objects, spaces and moods into words. A person on their device next to a statue of novelist Virginia Woolf. (Matt Brown/Flickr/Wikimedia), CC BY

You are sitting at your computer, interacting with a generative AI model like ChatGPT Image or Midjourney. You have a distinct picture in your mind, and you begin with a simple, general prompt: a chair in a cozy room.

The image appears, but you frown. You realize that to get what you want, you must elaborate, so you experiment with more descriptive prompts: Dark mahogany wood. Dim yellow lamplight. Late autumn dusk. You keep revising, trying to discover which words the machine needs and which words it ignores.