Quicksand Fans

This may seem off-topic, but please bear with me.

I recently discovered artificial intelligence-create art. The level it has reached is mind-blowing. For example: https://www.nytimes.com/2022/09/02/tech ... tists.html


Then I discovered a number of online sites where one can type in simple text, and an Ai image generator will create an art piece for you. I tried several sites and results are mixed, some much better than others.

But at one site I received this when I typed in "QUICKSAND GIRL"...




Isn't that amazing? KInda eerie too. If you want to play with it the address is: https://hotpot.ai/art-maker

The open-source Stable Diffusion has already made this type of generator obsolete. It's even efficient enough that you can run it on a home GPU. You can also improve the faces by running the images through GPF-GAN.

Fuck! It's a bitter pill to swallow when a few lines of code makes better QS fakes than I can.

redjak6t4 wrote:Fuck! It's a bitter pill to swallow when a few lines of code makes better QS fakes than I can.

Engineering these kinds of results is still pretty difficult - there are tons of duds I didn't post. Be assured, we still aren't at the point where anyone can get good qs fakes out of these AIs on demand. When the movie Tron came out, the Academy Awards denied the animation team from eligibility because "computers did it", when in reality computers helped skilled people make rotoscope effects at a faster rate. This is a similarly misunderstood advance.

Stable Diffusion also has a function called "img2img" where people have turned crude line drawings, along with the right descriptive text, into detailed images of the thing they (tried to) draw, based on their original lines. By repeatedly applying it with careful tweaks on selections of each part of the composition, an artist can create an original work, but with the detailing automated by AI. I imagine hand-made fakes could similarly be supercharged by this method. I haven't tried it yet, I'm still just messing with text prompts.

Oh, and all the fakes I posted before this were done by hand. This tech didn't exist yet.

BogDog wrote:
But at one site I received this when I typed in "QUICKSAND GIRL"...

Not bad for a machine. It rendered some nice sinking matter.

But...the girls...with those faces...they're nightmare fodder.

Nessie

These are the ones I messed around with. I used midjourney. Nothing compared to the OP's though. That's a whole other league.

This is something of what I've been talking about for years. Come back in a decade and you'll describe the scene you want and the computer will render the video for you. It will make crack addiction look like cotton candy.

Indeed. This area is developing at a lightning speed.

I've got decades worth of quicksand artwork and photography saved, maybe I'll use it all to train a fetish model.

So, in the future fantasy will merge with reality and we won't know if we are coming or going.

klib21 wrote:I've got decades worth of quicksand artwork and photography saved, maybe I'll use it all to train a fetish model.

Doing so is actually and then publishing the resulting art is a bit of a legal gray area. A lot of the images used in the training sets have been included without the artists' permission. Whether that sort of use violates copyright is still being debated.

But... since we are at this point now. Would any producer or artist here object to having their work used in AI training?
Personally I would not mind unless the resulting artwork looks too much like a rip-off of something I created.