Artificial intelligence is (rightfully) lampooned for all sorts of reasons, but credit where credit is due: neural networks really know how to make some freakishly beautiful art.
Or at least, that’s my opinion after seeing the recent surge of platforms like Midjourney and Google’s Imagen that are designed to swallow basic strings of text—“a painting of a flower,” say—and spit out such a painting right back at you. And if you haven’t any of tried those, maybe you’ve heard of what has become the most prominent player in this space, DALL-E, which can summon pictures of anything from a hyperrealistic avocado-themed teapot to a hyperrealistic avocado-themed chair, to a painting of a dapper bunny sitting on a park bench. But access to DALL-E (and the new-and-improved DALL-E 2) is currently waitlisted, which leaves wannabe artists like myself sitting on the sidelines.
Thankfully, the art-gods (or more specifically, a Houston-based programmer named Boris Dayma) has taken pity on the unwashed masses without DALL-E access and given us the next best thing: DALL-E Mini, which you can try for yourself. As Dayma puts it, the tool is a bare-bones, open source version of its namesake, that anyone can access and run freely in their browser.
Dayma has noted in interviews that the model underlying the tool is still training, which means the results Mini gives aren’t necessarily as polished as the OG DALL-E’s. It also might take a minute or two for the AI to work its magic (and you might need to resubmit that request a few times), but in the end, if you tell DALL-E Mini to imagine up “a buff picture of the Duolingo owl,” then yeah, the tool’s going to spit that out. And if you tell it to imagine up “the night terrors someone only experiences by passing out during a multi-hour Seinfeld marathon,” you get the images that follow here.