Generative AI & the senses
Image generation is the use of AI to create new pictures from a description, a sketch, or another image, rather than retrieving photos that already exist.
Type 'a lighthouse in a snowstorm, oil painting' and an image generator paints it. The picture is not found on the internet; it is produced fresh, pixel by pixel, from what the model learned about how words and images relate. Change a word and you get a new picture.
As of 2026, most generators use diffusion models, which start from random noise and gradually shape it into an image guided by your text. Earlier systems used GANs. The models learn from hundreds of millions of image-caption pairs, which is how they know what 'oil painting' or 'snowstorm' looks like.
Image generation is used for concept art, marketing, game assets, and fun. It also raises hard issues: artists' work in training data, fake photos of real people, and hands with six fingers. Many generators now add invisible watermarks so AI images can be identified.
A teenager types 'a cat in a spacesuit floating past Saturn, watercolor style' and gets four different paintings in under a minute.