It'll be good for accessibility too; if you can make a text-to-image model then you can make an image-to-text model, and now you have a screen reader.