Love it. The image and the aspect ratio. Did you try/notice the panorama modifier listed in the new menus?
No, haven’t seen that yet.
This time, I deliberately aimed for more alien landscapes from two of NASA’s Astronomy Pictures of the Day. Came out quite well; I didn’t even bother trying to evolve the first one. Second one almost looks like it could be from No Man’s Sky.
Come into my early 80’s wood panel basement for some classic AD&D. As you enter the 10x10 room, you see a “Skeleton With Treasure” as painted by…
Larry Elmore
Yeah, that looks Elmore-esque. I actually wanted to try a female elf to see if the AI would paint her in fur-lined hotpants and 80s mall hair but, sadly, the limitations of the AI to depict humans (or demihumans) held me back.
Clyde Caldwell
This one was disappointing. Especially because I earlier tried a couple other images as Caldwell that looked closer to his style. I’ll toss them up in another post.
Jeff Easley
This is not only a pretty cool image, it actually resembles Easley’s brushwork versus the other “realistic” TSR artists of the era such as Elmore, Caldwell or Parkinson.
David A Trampier
Shamefully, I don’t think the AI has any idea who Trampier is. I also tried the same prompt with “Ink on Paper” since most of his TSR work was ink illustrations but it didn’t come out any better.
Erol Otus
Likewise, the AI completely failed to capture Erol Otus’ very distinctive art style. Not even in the same ballpark. DALL-E, we’re waiting on you to fix this.
Here were some dead ends or failures that came out interesting in their own right:
Beholder (Erol Otus)
Fails to capture the art style of Otus but it does have a semi-bonkers look to it.
Beholder (David A Trampier)
Again, doesn’t really look like a Trampier piece but it does have a more cartoony look that made me think the AI was Google Image Searching some Wormy. Also… is it drinking a pint of beer? While a failure in the “Give me a Trampier art piece” arena, I like this image.
Beholder (Clyde Caldwell)
While completely failing to draw a Beholder (I’m not sure the AI really understood the concept though I started with Beholder thinking it would be simple but iconic versus trying to draw dragons), the brushwork on it looks much more Caldwell than the Skeleton picture
Female Elf Archer (Clyde Caldwell)
This picture quickly convinced me that trying to do an elf was a waste of time but the brushwork and skin tones in the center felt very Caldwell. Yes, I spent way too much time viewing TSR fantasy cheesecake art as an impressionable youth.
I wanted to test if the AI would recognize paper types. I chose a term that I knew from experience was almost certain to generate an interesting image, “yokai full-length portrait”. Then I tried “paper” types that came to mind: vellum, washi, shoji, papyrus, and pasteboard.
This is washi, vellum, and pasteboard (in that order). I love all three of them.
The resulting images came out very differently despite having the same seed, but you can still tell where some paper influences are. The washi with crumples, the vellum with “veins” and crumbling corners, and the pasteboard having the irregular shape and broad, bright coloring that you tend to see in real pasteboard art.
Here are papyrus and shoji:
You can definitely see some papyrus color and texture in papyrus, and shoji must be trying for a watercolor effect, but I’m not really impressed by the subjects compared to the first three.
The yokai are understandably always dressed in kimonos. I wondered if I could get other clothing types (Yokai in x full- length portrait). Here are tuxedo, Armani suit, Hawaiian shirt, swimsuit, and spacesuit, all with the same seed. That, of course, led to another wondering, so here are five nudes with different seeds.
(Here is a set more crudely done but with some funnier clothing.)
It isn’t doing an active Google search when it starts building an image, it works off a pre-existing set. I think it uses this one, which I think you or I could download if we knew how to train a baby AI with it.
There are others, too:
Today I learned that Yokai have only one nipple.
Sorry, I didn’t mean that literally.
I’m a little surprised that a hundred million images is enough for some of the things I’ve seen it do.
Who said it was limited to 100 million images? LAION-400M by itself is already 400 million images.
The fact is, there were and are already DALL-E forks on Github, and at least reasonably large training libraries (OpenAI is not releasing theirs), so anyone with the computational resources can be messing with it (and all the other algorithms). But that does not immediately translate into free-for-all web portals.
I didn’t start out trying for another Alien Landscape here but that’s where it ended up. 
I guess APoDs are really good for generating alien landscapes.
Pokemon Battle | Style of Lascaux
I mainly wanted to see if it recognized that as a style.
Curiously, adding tags for “chalk drawing | stick figures” made the pseudo-Pokemon look like fluffy 3D critters fighting on a piece of parchment rather than making it look more like a cave wall drawing.
Speaking of styles, Michael Whelan, Wayne Barlowe, and Boris Vallejo all seem to have influence.
I tried “Wrapping Paper for Mummies” and got some papyrus looking wrapping paper with hieroglyphics on it which was pretty impressive.
Also I think it has an astronaut in the corner but who doesn’t like astronauts?
If you share your artwork directly from Nightcafe instead of an image host, we have to click on the thumbnail to view, but it gives us a chance to give your work a “like,” and lead to more free credits for you.
More working on modifiers. “Japanese scroll” can get good results. “Style of the bayeux tapestry” was disappointing in this case, but it obviously does something and could be useful. “Mosaic tiles” works, but I was hoping for smaller tiles. “Tattoo” is pretty cool, especially since it is presented on a lump of disembodied flesh.
For “leather art”, I was thinking of the type of images that are stamped into leather belts, bracelets, and wallets, and sometimes colored in lightly. What I got were some really cool 3D-ish images that you really should check the album to see.
These are ghost rising from a grave, same seed.
Wayne Barlowe looks more like Inferno than Guide to Extraterrestrials, but I see it. Michael Whelan works. Jack Kirby seems to be a miss, with this one at least. Leather art is a fail on this one—instead of “leather art” style the ghost is wearing a leather jacket and showing a painting.
Here is some “Dall-E mini” (~2.4 million images?) ghost rising from a grave, leather art:
A while back I turned a frozen soap bubble into an image. The moon on that one was distorted so I evolved it and got this:
This is the original image.
Tonight I did another one, replacing the sky. (Modified source image on the page.)