When an A.I. model is trained to create images from text, it uses a huge dataset of images and their corresponding captions. The model is trained by showing it the captions, and having it try to recreate the images associated with each one, as closely as possible.