Ok, this looks super fun: “we’ve trained a neural network called DALL·E that creates images from text captions for a wide range of concepts expressible in natural language“.