BNevis
patreon
It's All Fun & Games (& Exploitation)... Before Skynet.
🕑 Added 2022-10-02 22:42:25 +0000 UTCIn case you missed it, I asked people's opinion on AI generated art yesterday on my twitter. And the responses were pretty much exactly what I was expecting.
While there was the kind of 'WHO CARES?! MORE TITS!' contingent, the majority of folks were kind of in alignment with what I thought on the matter. About how it'll be a neat tool once it fully gets there, but is no way a meaningful replacement to producing actual, bespoke art, & how also, it's still unfair to the artists whose work it's sapping as fodder for its source algorithm.
Full disclosure, I had started messing with it myself about a week or so ago (shortly after I finished that first Drama at the Soccer Game piece). I saw another long time inflation writer posting some actually pretty solid results using Stable Diffusion & was cautiously intrigued.
So I stuck my toe in & started just trying to do some generic backgrounds. At least that was the initial thought. Because let's be honest, expansion art focuses primarily on what's happening to the subject. That's why I hardly ever do backgrounds for most of my stuff. It's not that I don't like doing them. I enjoy doing backgrounds actually. A lot. They'd just a gigantic time sync if I did them every time. And the results for them were pretty ok. Definitely something I could run through a few filters & flesh things in a little better maybe.
But of course, the temptation was too great. So of course I had to start testing other prompts. And I do find a lot of the results kind of intriguing so I figured I would share them here freely with all to show the sort of things I was getting. But to be very clear, these images are not by me! I typed words into a text box, hit a button, & a robot combed a database to piece these together using a vast library of other existing art & photosk spitting these out. On the two bigger ones at the bottom, I kind of compiled & photobashed some elements from a few in the same batch together using photoshop, but what I actually did was still very minimal here. Like I was saying, these are just tests & beyond seeing what I could coax as a result, at best some might turn into a reference image for something I'd draw later down the line. You're welcome to use them for reference too.
You can make out some of the themes I was attempting. From 'Chubby Korean Woman in Turtleneck' to something more along the lines of visualizing 'Vin de Bleuet' to even some wild mixed but all interesting results for something specific as 'steampunk Eurasian blimp aviatrix.' Mind you this is probably the better 15%, out of even the stuff worth saving. There were plenty of grotesque body horror style images I saw no point in keeping. Not going to lie, it's hard to get an inflated shape when the computer barely understands what even classifies as a human shape, with four arms & each with seven fingers.
It's a fun tool. And I'll be honest, as someone with an addictive personality I try to keep very much at bay, there's almost a gambling high, clicking that generate button, waiting to see what drops next. How close you can get something to decent looking. Most of these, like 90%, were only from yesterday, & I stayed up into the wee hours adjusting seed values & going again.
Will I make a habit out of this? Nah. I'm not opposed to maybe fudging a background here or there. But I don't really think it's all that ethical. I mean that's kind of why I'm just posting this publicly. It's not MY work, just what I was able to pull from the samples. I'm someone that does fall more on the open source side of the argument of things, but I mean, c'mon. I have a damn patreon! I understand we're not in a system that allows that to happen easily. And if you want to share things freely, that's fantastic, but I don't think it's fair to have other folks just dub your material as 'open source' for you. Down the line, as kind of free market as it sounds, if this technology is to exist regardless, I hope living artists & estates, people still benefiting off recent art, are able to sign exclusivity or licensing or 'likeness' deals, or maybe offer mini-databases of their own to plug in modularly (?), to still maintain some ownership & compensation over of what makes them them. At that point, it'll make these tools a lot more ethical to use.
Until then, back to the drawing board.
Comments
Achernarus
I'm playing around with it myself, but I'm not seeing a "Restore Face" toggle in Dream Studio.
BNevis
Yeah I kind of like it as a time saver too insomuch as a preproduction tool. Time I used to spend looking up reference material, I can try tweaking in exactly what I'm looking for & maybe get what I'm after or maybe some slightly different results that spin it in a slightly different direction. It's nice for brainstorming like that. But trying to kick things out & claim it on its own is some sort of amazing original art is like what '3d artists' used to do in Poser3D in the 90s. It's not. I'm hoping to try using it for background soonish, just as a proper experiment. But I also know that whatever I pull from it, I'm going to still have to paint over & tweak it in. But even then, it's just compositionally helping something I've done from scratch, you know?
BNevis
To be clear, I really didn't do much. Of anything LoL Stable Diffusion has a 'Restore Face' toggle, & I guess two different algorithms you can choose between to dial it in better. But one of the two kept giving my back errors. From there, in the 5 minutes of research I did, it was simply dialing in things that accentuated better faces in the positive & negative prompts. I noticed I got some almost photorealistic ones there for a second, but they all turned out very skinny considering what else I had in the text box. I kept tweaking the descriptions to round the results out some more, but the faces got gnarly pretty quick. I think I got it back to where it was, but I was getting some good some super bad in the same batch.
WhosThisGuy
I think it's a good tool for inspiration, especially if you're stuck in a rut. The ginger woman in the white sweater about a third of the way down along with the woman beside her in the orange top look great. And then there's what I assume is something to do with the Victorian balloon craze which looks quite fun. But I don't think it ever can or should replace an actual artist.
Achernarus
How on earth did you get the faces to look so normal? Let alone beautiful?
