In-context visual learning, that allows to create more images based on given image