Not to brag or anything but I’ve had a pretty good day at Night Cafe. After a bit of a dry spell, I hit the top 5% (15th place overall (if I counted correctly)), with my Phoenix Rising from the Ashes; and enough people were impressed with it that I topped 1,000 followers for a sweet 250 credits.
I’m glad they changed the contest rules (they no longer allow any start images not created by the artist).
Nice!
I’ve been taking a break so I haven’t done it under the new rules yet. Maybe, I’ll give it a try. It was annoying me that all the top winners had their prompts hidden. I assume they still can do that.
Yeah, I don’t mind the hidden prompts, but some of the results I was seeing you just know were generated from a photograph and just tweaked enough to make it look original. The “Clear Words” contest, I think, was the breaking point. I threw dozens of points away on that run trying to get something close to a word. I did manage to get a stop sign that actually said, “stop,” but the winners were spelling out full sentences ferchrissakes.
Night Cafe now has the Stable Diffusion XL beta. It is more than one credit per image, but you get 10 credits for the first SDXL image you share. And once again you’ll need to learn a whole new way of crafting prompts.
And in other news, I looked to see if Playgroundai has SDXL yet. They don’t, but they have what seems to be an entirely new AI engine called Playground v1. You also need to relearn for that one, but at least it is free.
I made the comment in another thread…
“I want a painting of a vase of flowers carried by a camel sort of in the style of this artist but also in the style of this completely different artist but with more of a pastelpunk vibe and looser strokes but also it should be in a medium like spray paint applied to wet cardboard”.
…to which Darren_Garrison responded with an image based on that. I didn’t want to hijack that thread with image shenanigans but was inspired to run it through MidJourney. However, I did change the prompting to MJ’s “language” since my point was that I could make nearly infinite requests of an Art AI model, not that they’d necessarily be in natural language.
Spray paint on wet cardboard, by Arthur Rackham and Ghada Amer, a camel carrying a vase of flowers, Pastelpunk vibe, loose brush strokes:: --ar 3:2 --v 5
Spray paint on wet cardboard, by Jonas Wood and Norman Bluhm, a camel carrying a vase of flowers, Pastelpunk vibe, loose brush strokes:: --ar 3:2 --v 5
Spray paint on wet cardboard, by Lyubov Popova and Dan McPharlin, a camel carrying a vase of flowers, Pastelpunk vibe, loose brush strokes:: --ar 3:2 --v 5
Guess the AI isn’t great at spray painting at the moment. Oh, and artists pulled via RNG from a big list I have.
Playground AI has tweeted about the new model, but no details.
Introducing Playground v1: A new model to create images with an unparalleled spectrum of vibrant colors.
“We got brown… black… grey… you could make a c.2010 First Person military shooter video game out of this…”
It really does seem to be producing dark images. Without much variety or creativity.
I haven’t tried it but the irony of those example images was too much.
Well, huh. I’ve been using Bing’s interface for DALL-E and it will not draw me any depiction of the John Hancock building here in Chicago because
Looks like there might be some words that are blocked at this time
I’m guessing the “cock” of “Hancock”? Are you serious? Could their filter be this stupidly implemented, or am I missing something?
Just don’t ask about Scunthorpe.
That’s it! I was trying to think of what it was called when describing this to my data scientist wife: The Scunthorpe Problem.
Oddly, Scunthorpe does not trigger it with DALL-E via Bing.
But when I ask for a portrait of John Hancock, it asks me the Founding Father or the Insurance Company. So far, so good. I answer the Founding Father. It says “OK, I’ll try to create that” and immediately tells me that there may be some words blocked.
ETA: Huh. It will do Alfred Hitchcock, so I have no idea what is going on. But still no luck with “Draw me the Hancock Center” even after it looks it up and asks me if I mean the building at 875 N. Michigan Ave. (It’s actually not officially the Hancock Center anymore, being renamed to the address in 2018.) And if I ask it to draw me 875 N. Michigan Ave., I get some early 20th century building, not the skyscraper.
Oh, it absolutely could be that stupidly implemented. And probably still doesn’t land in the top 25% of stupid word filter implementations.
I once got an amusing warning from Midjourney for prompting a fairy in a giant cocktail glass because “giant cock” was prohibited.
Midjourney also released a /describe feature today which is its version of CLIP interrogation – feed it an image and it’ll give you four prompts you can use. Except it very often includes fairly innocuous words on the MJ ban list (flesh, chest, bare, etc) so you can easily feed it a pic, just spam the four result prompts (just need to click numbers) and get four flags on your account within seconds. Truly a well thought-out system.
The Hancock Tower isn’t the Handcock Tower any longer and has been “875 N. Michigan Ave” since 2018? Wow, I heard nothing at all about that one…
Well, it’ll give me Buzzcocks. So I ask it: “Is Hancock a forbidden word?” And it answers: “No, Hancock is not a forbidden word. It is a common surname and placename in many countries … [examples of various Hancock places and people] Why do you ask?” "Because you won’t draw a picture of the John Hancock Center. I get an error that says “Some words may be blocked at this time.” To which it answers: “I’m sorry but I prefer not to continue this conversation. I’m still learning so I appreciate your understanding and patience.
” And end of topic.
Weird.
ETA: Huh. It’ll do Herbie Hancock, though. But not John Hancock. Ah, but just “Hancock Center” seems to work. It’s the combination of John and Hancock that twigs it out, but only seemingly when drawing images (so I guess some filter on DALL-E’s side?)
If you don’t mind a bomb with handlebars like a motor cycle…
That’s awesome.
I’m currently at 30,000 feet over Arizona, generating silly pictures and giggling to myself in my seat…
I’m not sure if I’m being wooshed here. I have seen “John Hancock” used more than once as slang for a man’s penis for many decades now. Its filter is doing its best to make sure forbidden things aren’t slipped past it with euphemisms. Once you do manage to slip something by its filter, it probably will happily draw that body part it’s normally not allowed to depict. They have to be paranoid.
I’m not really aware of that, any more than any -cock name could be used as a cheeky euphemism. I mean, it sounds vaguely familiar, but nothing that’s in my living language experience among my peers or media consumption. A John Hancock in slang is a signature to me. But two points: it can just interpret it literally and ignore any slang meaning to it. If I ask for a picture of a beaver, it does t assume I’m talking about a woman’s vulva. Second, I’ve given it context: John Hancock Center. Plus it even asked me in one dialogue tree to clarify whether I meant the founding father or the insurance company when I simply asked it for a portrait of John Hancock, and it still rejected it.
It’s probably a case of a system with different amounts of smarts in various sections. Does it offer to accept feedback whenever you trip its filter?