DALL-E has been getting a lot of my attention lately. I signed up for the wait list to become a user and got an invitation a few days ago. DALL-E is much more capable than DALL-E mini, later renamed Craiyon, that I had played with previously. It generates images higher quality images, faster.
I’ve been using DALL-E like a bad graphic artist’s client, pestering it every few minutes for free concept art with the promise of “exposure” as its own reward. DALL-E doesn’t seem to mind, though. And we love each other. I ask it for whatever comes to mind, and it seldom disappoints me.
I decided to ask DALL-E to create box cover artwork for Mega Man, since why not. It occurred to me that this would be a good John Henry competition, pitting the new machine up against the mightiest artist to ever lift a colored pencil, to see who was better.
It turns out that DALL-E was more than up for the challenge.
I mean, none of these is exactly good, but all of them are awesome.
Most of these don’t feel like Mega Man, but they all have a pretty good Japanese Giant Robot vibe, which is just fine. Mega Man is a small robot, not a giant robot, of course, but that hardly matters — instantly, my imagination is fired by the idea of a 1960’s or 1970’s Mega Man anime imported to the US in the 80s alongside shows like Star Blazers, Mobile Suit Gundam, Mazinger Z/Tranzor Z, and Macross/Robotech, and Diaclone/Transformers… What a “What If?” to think about! So the nostalgia factor for this alternative Mega Man that never existed is powerful.
It makes me think that if I got back into game development, I’d use DALL-E to give me the initial inspiration, concept art, cover art, what have to get me going.
I have so many questions about DALL-E.
Why does it get the title wrong? Megan? Meggian? Megman? It can’t be hard to get the exact text out of the description I entered and replicate it in the image. Is this a deliberate nerf on the part of the developers to prevent users from creating meme captions that could be offensive?
Due to the way DALL-E can’t seem to get the words right for the titles, a lot of what it generates reminds me of this hilarious meme from years ago, which parodied corporate fast food logos by turning them into dada-ish nonsense. Which makes it unintentionally (or not) hiliarious and entertaining, but it also limits its usefulness for creating images I want to actually use.
It’s exciting to be alive at this time, getting to see these developments in AI. I don’t get the sense that DALL-E is truly intelligent, but what it does is impressive nonetheless. The images it creates are often quite good, and about as often are complete nonsense. But frequently this “nonsense” is either hilarious or entertaining, or has a superficial feeling of artisticness to it. If I knew a human had created some of these “artistic” images, I would be more inclined to ascribe creativity or meaning to them. But what when DALL-E does it, it feels more like it is holding up a mirror to our culture’s human-created art, as though it put the conceptual ideas of art and design into a blender, and reassembled them for us, not quite at random, but according to rules that don’t quite work yet, but still give the impression of… I don’t know, “something going on in there.” Often times the results it spits out give me the impression that I’m looking through a window into an alternate reality; as though, the many worlds theories of physics are true, and DALL-E is somehow extracting images of alternate realities from one of an infinitude of alternate universes in our multiverse. And from an information theory standpoint, perhaps that might as well be what it is doing.
When the images are purely visual representations, the impression of intelligence at work is more effective, but when the images contain words, it’s clear that DALL-E is just throwing shit at the wall to see what sticks, following rules of some kind, but not intelligently, not with deliberation or intent. We get nonsense words that are a jumble of letters that the DALL-E parser decided it thinks we wanted it to put into the image. And sometimes we get strange alternative glyphs, as though from some alien alphabet.
Much of the time the images DALL-E creates could be used to stimulate human creativity, to give a human creator a spark that they could use to work from, an initial inspiration. That’s a valuable thing in itself. And I’m sure the potential is there for it to continue to develop its abilities, and I expect that in time AI development will give us more capable AI. That seems almost a given.
Seeing what DALL-E would create was so fun that I asked it to create alternative cover art for many of my favorite games. There were many that I thought would be fun to share.
Legend of Zelda
Super Mario Bros.
Super Mario World
Ms. Pac Man
Teenage Mutant Ninja Turtles
Bad Dudes vs. Dragon Ninja
These are amazing, especially considering the amount of human effort it takes to create them — literally just typing in a few words and clicking a button. They are so fun to look at, to anticipate what DALL-E will come up with while the progress bar fills. Each one is as interesting for how it fails as it is interesting for how it succeeds. Each seems like an alternative universe’s version of the game title. The art style varies — some of it looks like the illustration on a children’s cereal box, a lot of it looks like the cover of a paperback book or movie, while others look like the box a board game is stored in. Tetris appears to have strong links with Rubik’s Cube, and here and there I can pick up hints of other influences. It’s especially fun to see when DALL-E “knows” the game title, and picks up on stuff that is recognizable from the game, and uses it in the images it composes.