Now I want to input entries from the D&D Monster Manuals.
Going for some classic D&D critters that aren’t well known otherwise (no dragons, trolls, unicorns, goblins, etc)
Stirge
Looks more like a bowl of beans. I know stirge comes from Greek myth but that didn’t help the AI nail it down.
“Hook Horror”
Manicured kraken threatening a swamp freighter. I originally did this without the quotes and got a bone fishing hook hanging of a wall with weird squiggles
Otyugh
something organic and …moist. Often with sphinctery, eyebally, or lippy parts
“Displacer Beast”
Well, there ya go. Definitely feline with extra limbs and tentacles. Granted the tentacles are coming out of the extra limbs but it seems to understand the concept.
Owlbear
Easy win for that one. If the face was better, this could be a piece of D&D art.
Displaced Beast
My tablet autocorrected Displacer but I liked this image of a water-walking tentacle monster stalking a ship in harbor anyway. Someone stat this guy up.
Like I said, lots of names.
Yep. One of the prompts is Hangman’s Alley.
I thought of that for my series of test images. It was one of the clear successes.
Castle on a seaside cliff | style of dungeons and dragons guidebook:
(As a total aside, in a moment of drawing a blank on what to try, I typed “castle on a seaside cliff | style of crumpled paper from the garbage”. It turned out to be one of my favorites.)
(And try “found objects” as a mod—it tries to make things out of various crap. (But I haven’t actually thought about trying “crap” until just now.))
Started with a bunch of gold-colored noise. I like how this turned out:
Evidently evolving a third time covered virtually all of the buildings with desert.
More D&D critters
Grell
The AI managed to nail organic & wet while completely avoiding any semblance of “floating brain with tentacles” to make me think it knew what it was doing.
Bulette
I could see that being a baby bulette. Cruising down a school hallway, as baby bulettes are prone to do. It seemed to know I wanted a shelled digging monster.
Illithid
Too easy. Though I thought using the real name would be harder than “Mind Flayer”
Froghemoth
Those are scuttering weird little frog things but they ain’t froghemoths.
Flumph
Floating disc with little tendrils? I think this counts. I like it hanging out in a garden.
Gelatinous Cube
I mean, yeah, you better get that one right.
“Rust Monster”
We got rust and we got a monster, but we ain’t got a rust monster.
“Carrion Crawler”
Not a weird caterpillar thing but, ironically, those spidery monsters look like they’re crawling over a decomposing grell. Full circle!
So I decided to revisit my mushroom monster/yokai photo now that I know a few more tricks. The things I was trying were giving disappointingly bland results and with three credits left I decided to abandon mentions of mushrooms alltogether and make some old reliable “parade” shots (which are usually done in widescreen without a photo, but it woukd be an interesting experiment). So I did yurei parade, yokai parade, and robot parade, and ended up with sad cyclopses with elaborate hats.
(Full settings x | full-length portrait | tronie | wanted poster | symmetric face | hyperdetailed colored pencil drawing | style of Mark Ryden | diorama | 8k resolution | hyperrealism | volumetric lighting | photorealistic,)
I thought the third one was best in almost every way: the face, the hat, the background—but I liked the skirting on the mushroom cap that was in the first two but not in the third, so I copied that over. Also, after doing an AI upscale there was a white line cutting through the “eye”, and had to go back and darken a single-digit number of pixels to fix that. (But I also shaped the white of the eye a little.) I love the final result.
Here is the upscaled version:
This brings to mind another happy accident with the same photo. In my first round of experiments I noticed in the AI-blurred starting images the base of the mushroom looks like a very appealing dollop of whipped cream. So I tried a prompt about a mushroom yokai standing in whipped cream. I got an eldrich horror in a frilly pastel dress.
Okay, here is a sampling from my image tests. I type only “castle on a seaside cliff | style of x”. Renderer is coherent, aspect is 16:9, seed is 466340 (for no reason other than it was the random seed of my first test). I have lots more, some successes, some failures. I found that the AI apparently thinks a castle has black shadowy areas at the top. Some renders hide it well, some not.
(The image that shows up in the preview is Ralph McQuarrie.)
For some failures, here are Boris Vallejo, Cat in the Hat, How the Grinch Stole Christmas, Nightmare Before Christmas, Pink Panther, and red blue anaglyph.
I’ve been thinking about this for a while now. Today just happens to be the 42nd anniversary of Mount Saint Helens killing 57.
I like the embossed look of that.
I continue to dig about the internet for artist names that might make a significant mod. Yesterday I found Kenny Scharf (plus more). Kenny’s name produces extremely colorful, intricate images.
The colors reminded me of temari, so of course I tested the AI:
And it knew them. (Even though one has hair and the other is either eating or giving birth to a small one.) But I wanted more than two or three, so I tried “several temari lying in a grassy field”.
And got four. But two and three-quarters of them are made out of grass.
I store away images that I think might be useful, and grabbed a few that I thought might hint at multiple temari. Most of them didn’t, but still had interesting results. (Part of a seashell produced a series of shelves full of balls, but not detailed colorful ones.)
(One of the other names I found that worked yesterday was Frida Kahlo. But Frida Kahlo painted Frida Kahlo, so the mod makes things into Frida Kahlo, which gives it limited utility. “Hey, who wants to see a beloved female POC artist, but as a zombie?”).
Every once in awhile I go and mess around with this one some more trying to get both a recognizable joker face with a harley quinn face. But it’s so hard to get faces at all, I fear I won’t ever get it.
I really like the room and the general ambience even in all the bunches of iterations I’ve tried.
Study your progress images. Often they seem to be headed in a good direction but veer off wildly. For instance, watching this image form, I expected it to become a tattoo between her eyes (I get tattoos pretty often with the “wanted poster” mod). It did not.
You might find some small flaw in an iteration 50 image that you can edit and reuse as a start image.
(I mentioned once that in my own head I call these iteration 50 images “shades”, because they remind me of shading maps used in some CGI programs. I think the shades can look pretty on their own, and keep lots of them. Feel free to use any of them for your own projects.)
Gave me the idea to generate hyperdetailed charcoal drawings in the style of B&W photographers. Here’s Ansel Adams:
I have an Ansel Adams castle test. It suffers from the annoying black blocks I mentioned.
Good tip. Will give that a try in the future.