Well, the first Tokyo has Mt. Fuji in the background, anyway.
Wanted some oddly-colored plants (think of the red plants on the planet in the nuTrek movie.) The AI apparently can’t grasp plants that aren’t green.
All done via Artistic with no additional prompts or modifiers
Tree with Red Foliage
Japanese Red Maple
Tree with Blue Foliage
Colorado Blue Spruce
Seemed to resist the Tree with Blue Foliage. Looks more like a tree with green foliage and blue fruit. I’m amused by how it gave me an accurate looking, but tiny, Colorado Blue Spruce.
My Japanese Red Maple looks like it’s being sold in a garden center.
I wanted ground cover. Try for non-green grass.
It does look tiny, but I’m not sure why. The surface it’s growing out of doesn’t look like anything I’m familiar with, so why are we mentally assigning a sense of scale to it? And in fact, a sense of scale so strong that it’s overriding the sense of scale implied by the trees that we do recognize?
I guess I assumed it was a cliff face (since that correlates to the upper left one being in perspective) but really I just meant “tiny” as a comparison against the canvas size since I only asked for a single object.
Field of Red Plants
Field of Blue Plants
Red Autumn Moor Grass
Blue Hosta
It does seem a little more resistant to non-green herbaceous plants though I’m guessing that it has less frame of reference for what ornamental grasses and perennials look like. The red field isn’t awful but the blue field completely missed the mark. Bad AI, bad.
For giggles, I tried Imperata cylindrica (Japanese Blood grass) and it was… interesting.
It recognized that I wanted a plant but gave me one not remotely like the grass I asked for. I assume the round shape came from “cylindrica” but the overall effect looks sort of like a Pitcher Plant (Nepenthes species)
Jungle of Red Plants | Red | Red Foliage | Red Vegetation
Jungle of Blue Plants | Blue | Blue Foliage | Blue Vegetation
On the Blue one, I set a second prompt of “Green” to -1 but it doesn’t seem to have helped any. I also tried a final blue, adding in “Alien Rainforest” and increasing my exclusions -1 line to “green leaves | green vegetation | green” but got a nearly identical result.
A few of my latest creations:
This last one I evolved from a picture of myself before I cut my hair, with a lot of modifiers. I’m gonna hold on to it for potential avatar material:
Okay, some of those prompts might help me, if they work on coherent. (I needed something non-green to go in the background of something mostly green.)
Spawned a couple of haunted forests off that concrete wall:
That does look like a face by Charles Addams but I was originally trying for the whole family, not just giant Cousin Itt with a haircut.
It is so gratifying when an image works pretty much how you wanted it to on the first try. I used a photo of an unripe fruit stalk from one of my Amorphophallus konjac plants to make a monster covered with eyes. Ideally I would have liked more detail on more of the eyes, and all the same size instead of the one big one, but those are minor complaints.
I decided to try a video, which meant going Artistic. I used the 10 Credit Maneuver (open an incognito session just before 8 PM, get 5 credits, wait until just after 8 PM, get 5 more) and generated it with the same settings that I use in coherent. But not only did the still image turn out crappy, but the video? While Coherent knows how to use “hyperdetailed colored pencil drawing” as a style, Artistic stuck a bunch of freaking colored pencils in the video, which is both highly disappointing and kind of hilarious.
Just thought on another style to try: Google Deep Dream.
It works. Not quite the same, but definitely trying.
Summary
Castle on a seaside cliff | 8k resolution | hyperrealism | volumetric lighting |photorealistic | detailed painting | style of google deep dream
(Coherent)
(And yes, I see that the official name is DeepDream and I used Deep Dream. Still worked.)
I tried that same prompt last month (wow, have I really been playing with this thing that long?), but in oil painting style:
Oh my god you guys, style of South Park definitely works as a mod. Here are a few yokai parades and yokai portraits with soSP. (All have hyperdetailed colored pencil drawing, too.) Besides flattening the image, it tends to add some sort of red cap somewhere in the scene. Check out the tiny Kenny in the top middle image (which also has origami).
Next up, diorama is a heavy-hitting mod. Here are yokai parade, yokai parade, seashells on a sandy beach, alien landscape, Paris at night, and castle on a seaside cliff, all with 8k resolution | hyperrealism | volumetric lighting | photorealistic | diorama.
And here are Children in halloween costumes trick-or-treating | hyperdetailed colored pencil drawing | style of south park | diorama | 8k resolution | hyperrealism | volumetric lighting | photorealistic | deep focus Especially check out the fangs on the kid in the top left image.
And these are more yokai parades, same settings as above except “yokai parade” went where “children in Halloween…” was. (Also, some are origami.)
And lastly combining the two is Yokai in halloween costume | full-length portrait | hyperdetailed colored pencil drawing | style of south park | diorama | 8k resolution | hyperrealism | volumetric lighting | photorealistic | deep focus | golden hour
It is pretty amazing that the AI didn’t look at that prompt and say “Screw you guys, I’m going home.”.
For other shows, I saw what might be hints of influence from Bob’s Burgers and King of the Hill (maybe Linda’s hair? maybe the guys standing in the alley) but if so they didn’t do enough to make me dig deeper.
BTW, for testing settings, along with yokai I’ve started doing “castle on a seaside cliff” with a specific seed. I’m building a library of those.
Neat! Definitely is. Tricky to work with though. I’ve gotten some errors while trying to evolve into a diorama.
https://creator.nightcafe.studio/creation/AS2m8aWfTsfQOYxlC7KC
Meanwhile I just found another one: wanted poster.
I hate it when you go in with some grand idea and you experiment with it, evolving it this way and that, doing one’s best to carve out something … and you just get a bunch of non-artistic crap. This is my attempt at Satellite Overlooking a Dying Planet. This is like the 10th evolution. It sucks.
I did some successful things, like using the happiest, shiniest descriptors but using the mos horrific artists:
And vice versa:
Some others I like: