Switching out some prompts can help.
https://creator.nightcafe.studio/creation/J4efdBqHEmAZyw3DScmg
I was trying for your dying planet as seen through Enterprise D’s viewscreen.
I threw in that “hypermodernism” modifier without really knowing what hypermodernism is; that might be a monkey wrench too.
I spent time this week trying to coax photos I have taken of maypop flowers into a creature with a colorful frilled collar around its neck (ideally organic, I would have settled for clothing) but never succeded. Ended up with some interesting alien/fantasy flowers, though.
When you Coherent users do an image, where do you set the accuracy? Default or crank it up?
Spaceship, Style of Intellivision
Well, not really in the style of Intellivision but I think that’s supposed to be an Intellivision controller with the keypad and disc. Maybe Intellivision is too obscure to be a style.
Spaceship, Style of Atari
That looks even worse. I don’t even get some wood grain or an Atari joystick.
Spaceship, 4-Bit Graphics Style
Eeehhhh… maybe? I mean, no one would look at that and say “That looks like 4-Bit graphics!” but maybe it’s intentionally trying to be simplistic?
Spaceship, 8-Bit Graphics Style
You could charitably say that the background is made to be in the style but then you have that weirdly detailed spaceship.
Spaceship, Pixel-Art Style
I want my money back.
Default.
I mentioned trying Castle on a seaside cliff in various styles. One was Legend of Zelda, which I thought came out pretty well.
Then I did Donkey Kong
and Super Mario Brothers.
Kong and Mario don’t try for the low resolution, but I still think they get some of the spirit of the games right.
This is one of my favorites so far. For the record, I did crank the weight on the accuracy. I took this photo of my dog, Bob, and my uncle’s dog, Murray, which I thought looked like two dogs posing as a folk duo in the 60’s, and I created this … one evolution:
I remember talking about the complexity of a prompt. Now I’m doing prompts with a dozen parameters.
Thinking about how unsymmetrical faces d to be, it hit me: why not try adding “symmetrical face”? So I did. It doesn’t completely solve the problem, but it lessens it, especially when paired with “wanted poster”. (Except for the odd disaster where it does nothing at all.)
I tested it with some yokai portraits, and one seed produced an especially interesting result, so I decided to see if it could produce reasonable humans. So I thought of five random one-word stereotypes to replace “yokai”: priest, rabbi, imam, Crip, and princess. All of them turned pretty well as caricatures.
It is weird how some seeds produce very similar images each time you run them but others make something completely different each time*. For instance, I love the art of Mark Ryden, so I’ve been working with his style (the AI knows it reasonably well). One seed/prompt tends to turn out a variation of this image (the first is my favorite—it looks like she is holding some evil spellbook or evil boardgame, or maybe Pandora’s Evil Box):
Except sometimes it tosses out something different, like this:
(Possibly a playing piece for an evil board game?)
If you want to play with them, the link for the caricature producer:
the link for the Ryden image producer:
That DOES look a lot like the cover art from Atari games, though.
That concrete wall is really useful for charcoal drawings.
Feel free to download a copy and try it for yourself. Here it is again:
From one of my noise files, converted to greyscale:
I’ve been running another little experiment. I took an initial image, just the prompt, “Swamp Thing” and ran it through artistic. Then I decided to use that as the starting point to see what random artists from the modifier menu would do with it. For each artist I just added, “by their name” and then ran that Swamp Thing image through Artistic, then Coherent, then back into Artistic. I also ran the original Swamp Thing image first through Coherent then through Artistic, essentially giving me five results for each artist to see which comes out best. The results are interesting. I published the ones I like for each of the three artists I’ve done so far (Leonid Afremov (never heard of him), Kandinsky and Rafael Santi) though the one I like the best was a different iteration for each.
Original Swamp Thing image:
Afremov (Coherent):
I’m going to try different subject and different artists too.
Kandinsky (Artistic, Coherent, Artistic)
Rafael (Artistic, Coherent):
About landscapes, earlier I mentioned doing experiments with castle on a seaside cliff | style of x? Well this image came from nothing more than that plus “drone photography”.
So far I haven’t been able to create anything that spectacular again using all sorts of “this thing is up in the air” hints, but you might want to try.
I thought about (but didn’t) trying “swamp monster” in my quest to find creepy creatures to create that disguise the fact that the AIs are bad with faces. I know I mention yokai a lot, but monster and troll produce useable stuff, too (along with a few other mythical creatures). Things like elves and faries (which you expect to look rather human and even beautiful) are bad choices as well. My most recent try that worked was *yurei", Japanese for “ghost”. I’m thinking of digging into the mythology of other cultures.
Okay, you pushed me into trying one I had been thinking about. There is a monster in Asia that looks like a normal person (usually a woman) during the day but at night the head flies off (with the internal organs still attached and dangling below) to do various nefarious things. It has lots of names, but I tried “krasue” and also copy/pasted in the Thai word. The AI seems to have some grasp of the concept, but not enough to produce a successful image. (Some interesting failures, though).
I hate to keep posting, but I notice on two of those figures it looks like the AI is trying to spell out the word “krasue”: once in black on a red knit sweater, and once in florescent green just below a mouth. I had something like that happen earlier—I had just read about a new restaurant opening nearby called the Cotton Duck Cafe, and, being at the computer, wondered how the AI would handle it. Would there be a duck made of cotton? But I got this:
Anyone else ever have attempts at words from prompts show up?
Penanggal | Horror (Artistic)
Sort of an Asian woman monster head with no body and a seemingly severed neck. No dangling organs but it seems to have an idea of what we’re talking about/
Penanggal | Horror (Coherent)
Angry Asian woman in a work polo in what is perhaps the world’s worst physical rehab clinic.
I thought it was a manananggal? Maybe that’s a different demon? After 3 tries, nightcafe came up with one that did not have a lower half:
