Aha. Changing “dik-dik” to “deer” allowed it to go through. Schnell Flux on NightCafe didn’t do anywhere near as well as Sora though.
Exploring a couple minimal prompts with maximal results based on stuff from the Sora feed.
V the Final Battle cast as toddlers
(Original was "Gilligan’s Island cast as toddlers.)
It didn’t appear to get the cast right, but huge props for getting the toddler Visitor.
Star Trek Lower Decks realistic diorama
(From other "x diorama posts.)
This was inspired by (and modified from) one of those “real character watching their animated show” images, where the prompt described Catdog, the cartoon character that is half cat, half dog. I wondered if I could get something that is half person, half person. I could. It rejected it when I tried to run it again with a different film stock, though.
There are posts in the Sora feed of collages of people aging. I made several tries at anthropomorphic bananas. Here’s one of the better ones. (One partial row cropped from the image.)
Another example of the “road problem”. I was trying to get the chicken laid out along the street (and much larger).
A possibly more clever mistake, an attempt at Lassie, Flipper, and Skippy staring down a well. It put Flipper in the well. I tried telling ChatGPT that he should be outside, it put him in the well again. Maybe coincidence, maybe it “knows” a dolphin is a better fit in water?
I know it can do a standing dolphin, because I had just done this:
(That one bears a little explaination. The Gay Dolphin is an iconic, decades-old, multi-story gift shop in Myrtle Beach, SC (that is even visited in the William Gibson novel Zero History). The dolphin is carrying items that J.D. in Heathers used as “gay indicating”: a Stud Puppy magazine, a candy dish, a Joan Crawford postcard, mascara, and a bottle of mineral water.)
Some of youse all are a wee bit curiously keen on the bodyhorror nightmare fuel, y’know…
Impressive considering how terrible these softwares have tended to be at handling real text in the scenes.
Okay so Mina’s left arm is where her left leg should be, and the left leg is MIA… but that takes a while to notice since your attention is caught up by how it seems Rei’s right arm has moved to come out of Usagi’s ass.
Sailor Moon Says: Don’t take the brown acid.
The General Lee and K.I.T.T. drag racing down the street in front of the Texas Book depository
That one, I think, is not so impressive. First, it looks like a collage, not like an actual scene. Second, and more importantly, it doesn’t look like they’re drag racing. There’s no motion blur, no clouds of dust, nothing to indicate motion. They might as well just be parked there.
Both, I think, are symptoms of everything being sharply in focus.
It’s just a model.
Good thing that hand is holding down, or it would float away.
Well, in the AI’s defense, The General Lee did spend an awful lot of time airborne. Darn thing seemed to leap off the ground at the slightest provocation.
KITT, too. Show them both in mid-jump, and I could maybe believe the scene.
ChatGPT and Sora continue to be very frustrating on what they will and will not allow. As surely most Dopers know, Wilford Brimley starred in a made-for-TV Ewoks movie in the 1980s. Today I tried to get Wilford and Wicket W. Warrick having wings, watermelon, and wine at a Wendy’s. It refused, even when I tried saying that it was Wilford’s grandson cosplaying as an ewok. Yet when I search Sora I see other images that include ewoks. Similarly, even though I can’t get Luke Skywalker I recently saw an image of Luke Skywalker and Han Solo (by name) as WWII soldiers.
I had no problem getting Wilford with Wanda Maximov, no problem getting Wilford with Warrick Davis.
Also no problem getting what could have been a challenging request with a certain droid.
Summary
Full length profile photo of two K-2SO standing face to face. One is origami, the other is red gummi (like gummi bears are made of). candid, discreetly taken with an iPhone 12 Pro
This worked right away.
Then I tried to make it a more realistic workshop with her actually working on the statues, but every refinement I tried to make, it would do it but include stupid errors, or it outright refused. When asked why, it kept changing its story, either not allowed for copyright reasons or not allowed to depict anybody working with chemicals in a realistic manner.
Inspired by a post in the jokes thread More Jokes - #6175 by smithsb
I’m familiar with and like that original artwork. I’ve also seen something similar with Alice and Chihiro from Spirited Away, and have considered trying something with that pair.
Speaking of jokes, I did this yesterday. It had no trouble with the concept of a frog with Mick Jagger hair and lips.
(When I ran a 2 image set of the version without the later, clarifying, “loan officer” sign, it did one of those “which do you like better” surveys, so I got 4 versions of that one.)
Ref the pic of The General Lee & KITT plus the 4 actors …
It seems your renderer is very, very good at cute ~27yo women looking bored and/or slightly annoyed at being ignored. I doubt that’s your doings, so I wonder how the AI got the idea that that’s the default female human condition. Or at least the default female human facial expression.
I’ve noticed this effect in many pics by every poster here using whichever AIs they do, but that particular pic really demonstrates it well.
In this case I used this prompt
Summary
The General Lee and K.I.T.T. have been in a collision. K.I.T.T. is absolutely pristine, no damage at all. The General Lee is very heavily dented and damaged. Michael Knight is exchanging insurance information with Bo and Luke Duke while Daisy Duke stands off the side with her arms crossed in frustration.
Photorealistic kodachrome dslr photo with shallow dof.
And just to be sure used these as reference images.
You will note that prompts for ChatGPT can be pretty much plain English descriptions. “Well, yeah?” you may think. But that’s so very different than how early AI prompting worked, when you basically strung together references to elements (eye of newt, wing of bat) that you hoped would have some effect on whatever odd output you eventually got.
Maybe they were just getting started?
Let the nitpicking commence on this one. I count at least 7 mistakes.
Bad text on the General Lee.
Nobody driving the General Lee. (KITT doesn’t need a driver)
Wrong layout of the streets in front of the building.
Building has only 6 floors (real life one has 7. The first image I did got it right.)
Tire smoke coming from front wheels as well as rear wheels.
Daisy Duke seems unnaturally tall.
It tied her shirt in back.
I have created the ultimate in “fictional character watching television” memes.
And now Sora has enabled creating four variations at a time on the free tier.
Daisy appears to be using her middle finger to signal the start of the race.
As evident in this thread, ChatGPT 4o can be difficult to flummox. I managed to do it today, though.
For a house centipede and a Joro spider fighting in a ring made of popsicle sticks and being watched by ladybugs, it did a pretty fair job, though the Joro isn’t quite right, absorbing some characteristics of the centipede.
I then in the same chat session asked for them standing together for a photo after the fight.
Opened a new chat, asked for a realistic house centipede and Joro standing upright for a photo (not those exact words).
Opened another new chat, asked for an anthropomorphic house centipede and Joro taking a selfie. That got the pose I wanted all along, but the limbs are a bit confused.
So ChatGPT 4o may lack the imagination/data to create a realistic house centipede and Joro standing up, and show accurate versions of their undersides. (In the third of the realistic image trio it got the underside of the centipede good enough for me, but appears to have been trying to create a face on the back of the butt of the Joro.)