They're currently all generating on the fly, just from summaries of what is supposed to be in each image (and with a lower-quality model) instead of the full description of objects, style, etc and on a larger model. :)
But yeah, you can currently go anywhere and have it generate on the fly! :)