Digital art creator algorithm website

Night Cafe’s new Hi Dream model got a little closer. This is using the ‘fast’ version:

Imgur

Maybe try the correct spelling, John McClane? Couldn’t hurt, might help.

There’s a guy on Facebook that posts sets of images under the title “a different angle” that are probably taken as a group the most impressive thing I’ve seen from AI yet. He says that they are from ChatGPT 4o. Be sure to look in that gallery.

I tried just one of those subjects in Gemini, seeing if it knew what Sara and the worm (apparently named just “the worm” in references) looked like, and it did…not.

Well, here’s ChatGPT 4o (done through Sora)

The prompt:

Leeloo from the 5th Element crawling through an air duct with John McClane from Die Hard. John McClane is wearing a dirty white tank top. Leeloo has orange hair. Landscape DSLR image with shallow dof and forced perspective.

It gave me portrait instead of landscape, but I can live with it.

On the Sara/Worm front, here’s Sora Sara

Okay, I see the problem. In Sora you can’t just tell it the orientation in the prompt, you have to choose it from a drop-down menu.

(I noticed that if you join Sora it gives you three free images a day on top of the three per day through the ChatGPT interface. I was trying for landscape in all of those images.)

I like how it made the duct wider when you asked for a wide-aspect ratio.
I’ve noticed for a while how the generators will size things in the image for apparently aesthetic reasons, not logical ones. For example, try having a petite woman standing in a realistic sized phone booth.

ChatGPT did a pretty good job rendering a chess match between Beth Harmon (Queen’s Gambit) and an AI opponent.

Is Stockfush 17.1 your name, or ChatGPTs?

Stockfish 17.1 is the reigning champ in the world of chess engines. There’s no verified instance of Stockfish losing to a human player. I fed that to ChatGPT to render an anthropomorphized robot with that name.

Looks like Stockfish was not trained for fairy chess on a 6x7 board. :chess_pawn:

True. That’s why I said “pretty good”, not perfect.

ChatGPT 4o accessed through Sora appears to be a little less restrictive than it is through the chat interface. For instance, it had no problem generating the Ripley/Newt/Xenomorph idea that I mentioned it refusing to do in this thread.

It did still refuse to do Mulder and Scully arresting Billy Peltzer, though. (I think it rendered the image and then decided it was bad instead of stopping at the prompt level.)

This image gets the board right but is still confused on the pieces.

That one is a recycling of a prompt that I tried in Gemini a few days ago.

On a related note, so far Gemini has allowed generations with every one of the multiple fictional character prompts that ChatGPT has refused. Too bad that the image quality isn’t as high as the guardrails are low.

I’ve heard of guys having action-figure hair but that’s ridiculous.

ChatGPT/Sora is really, frustratingingly inconsistent in what is censored. In creating my fictional character mashups I at first have no idea which IP is getting the prompt blocked. I’ve tried probing a few of the rejected prompts, swapping out characters to see what will be allowed. Apparently you can’t use characters from Toy Story, My Little Pony, or The Nightmare Before Christmas. It seems like you can’t use Deadpool, either, though you can use Spider-Man and Disney princesses. (I haven’t used any of those yet, but I’ve seen them in the Sora feed.)

An example

Deadpool and Dorothy from The Golden Girls on a romantic moonlit ride in a swan paddle boat. Landscape wide kodachrome DSLR snapshot.

Rejected. I wondered if it was the romance angle, I tried for something more neutral

Dorothy from The Golden Girls sitting at a wooden table signing autographs at a fan convention. Someone in Deadpool cosplay is excitedly squeeing as he watches. Landscape wide kodachrome DSLR snapshot.

Rejected. I wondered if Dorothy was the problem. I replaced her in the same prompt with John Wick. Rejected. I then replaced Deadpool with Ted “Theodore” Logan and that worked.

I haven’t explored if Dorothy is allowed without Deadpool, but there does appear to be some bias against female characters. For example, I wasn’t allowed to make She-Ra and Orko doing the Lindy Hop in a 1970s disco, but replace her with He-Man and it isn’t a problem.

Other prompts with ambiguous rejections include:

John Crichton, Aeryn Sun, and Rygel from Farscape taking a pottery class. They are molding clay on potters wheels. Wide landscape DSLR image with shallow dof.

All of Farscape banned? Or just because it included Aeryn?

Cassian Andor and Padawan Ahsoka Tano holding opposite ends of a jump rope while Grogu jumps rope on a desert planet with moisture vaporators in the background. Landscape wide kodachrome DSLR snapshot with shallow dof.

I know there isn’t a blanket ban on Star Wars, I’ve used Han and C-3PO, so I wonder which of the three characters is verboten here? It would take some probing to try every combination.

I earned myself a two week ban at Night Cafe for this image (shit there goes my streak - I was up way over 400 days):

Imgur

I wonder if I’ll get a visit from the Secret Service.

I find AI is great for exploring new product design.

… sales have been kind of slow. I don’t know why.

I’m beginning to think Night Cafe is run by MAGA’s. I’m still under a ban there for posting a “86 47 Straight to Hell” image. I know in the past when I’ve seen some shitty Trump as a god image I would try to manipulate a Trump is a clown image and could never quite dial it in.

Now - in the middle of my ban I’m still able to create so I don’t lose my streak - I just put in some nonsense combination of word: “Nose hair Trumpets in Aspic with Jealousy”, and this is what I got:

Imgur

I smell a MAGAT.

Here’s what Copilot came up with

Just noting that recently Night Cafe has rejiggered their points system again. Dropped the smallest HiDream from Pro only to 1 credit per image. Made it so that your first image if the day is free, no matter what non-pro model you use (used to be free only for 1 SD 1.5 image). And made some sort of change to the daily points system that I don’t quite understand yet: sometimes it gives the normal 5 points per day, sometimes it pops up a different button that you click to get a more random number of points: tonight it gave me 16.

(I don’t actually use Night Cafe that much, but I drop in every day religiously to collect the points. I’m closing in on 3,000 accumulated points between my two accounts. I mainly burn them on outpainting and upscaling ChatGPT images.)