AI art generator

All of these AI art generators are completely derivative, but it's still uncanny how many good results you can get through not that much trial and error. I'm pretty sure that this is going to be subject of litigation sooner or later, because it's basically profiting off thousands of artists' collective works that it's using as a base to generate derivative works from. Obviously it's not a problem for the art that was contributed freely, but I can't see the artists who don't wish to give their art away for free being happy with this.

 

What they generate is based on the database of existing work only, there is no originality in the human sense, the work produced stems purely from combining and manipulating the existing pool of art. You can easily produce art "in the style of artist X" that will contain recognizable features and feel like a piece of art made by Royo or whoever. So it's just very good at seeming original, but the originality comes purely from incorporating dozens of thousands of pieces of art, recognizing and duplicating patterns and rearranging them in random forms so that the result is different every time. Naturally I'm simplifying here, but that is the basic premise.

I wouldn't be so sure about that. Sure, a lot of art could be considered derivative, but no human relies solely on a database of existing work to produce new art like the AI generators do. The existing laws in this respect were made taking humans into account, not AI.

Once the AI generated art becomes widespread, I can easily see recognizable artists arguing loss of income when anyone can generate art that appears to be made by them, but the only one profiting from it is the AI generator company pocketing the money while churning out derivative works by thousands, devaluing the work of any given artist and robbing them of the profits from being able to sell their works for a price and in the quantity that they determine themselves. Obviously, arrangements where an artist would allow this would be possible as well, but they'd need to be compensated for it or allow their works to be used freely.

I won't speculate where you'd draw the line, I expect this will be determined in lawsuits in probably not so distant future. The music industry has relatively clear standards to what extent one song can mimic/sound like another before it infringes on its copyright and I don't see this being any different with visual art. And the boundaries in music have been clear for decades, so this is nothing new. A bunch of incidental or "incidental" copyright infringements happen there all the time and are handled via lawsuits, so I expect that the situation will be pretty similar with AI generated visual art as well. Behind every AI-generated piece of art is a long list of which sources the generator drew from to generate a new image and to what extent and how, so if anything, it'll be much easier to determine the extent of infringement.

 

That'd be the hard way. A court ordering the AI generator company to reveal the source artwork and the extent of its usage behind every generated image would be much easier.

Well, you know that this argument never really stopped being used anywhere. I'm sure it's going to come up here as well soon enough. What the European courts might determine as sensible vs. what's ruled in the US will likely be quite different.

 

It sounds kinda fishy to me that they wouldn't know exactly how their system works. Don't see how they could improve it if they didn't. Given that it all boils down to maths, surely the exact numbers behind every output should be possible to generate as well?

 

Gave it a try myself, with a few search terms. Just based on what I saw, it's hard for me to see original artists losing sleep over the so-called "art" it spits out.

 

You'd be surprised: https://www.pcgamer.com/ai-artist-who-won-competition-says-art-world-is-in-denial-about-the-tech/

There's a bunch of these and I've seen hundreds of great pieces of art generated by them posted by users on a few different forums I frequent. There's a bunch more generators, e.g. https://creator.nightcafe.studio/ and others. What Midjourney can generate (currently closed beta) is just scary.

You can easily get professional-grade art out with a bit of trial and error.

 

Denethor is one of my favorite characters of the not-quite-heroic mould. A morally grey and complex character amidst the white vs. black tension during the War of the Ring. He would be a Shakespearian-worthy main character in any story other than LOTR imho. Came across the below AI-generated picture, not quite how I visualize him personally. But DAMN that is an impressive depiction. Apparently using just the right search terms, which can be numerous, to get the result you want is now becoming a real skillset unto itself:

 

This is a classic problem with AI: Denethor has 5, possibly 6 fingers. For some reason AI still doesn't give 4 fingers to a humanoid portrait! (The knuckle of the sixth finger is just visible beyond the 5 fingers if you look closely.)

Bruh, I want 1 thumb and 4 fingers so that I can claim gratification in showing the middle finger to someone

Also, if that is not a sixth finger then his little finger is really long -- longer than the aforementioned middle finger in fact.

Note also that although the orb/bubble in his hand appears to be translucent, the end of his cloak's 'pin' (dunno what to call it) is invisible behind it. In "real" art that would have not been the case. You'd see some vestige of the end of that thing behind the bubble.

Yes, you can crop it out and only use the very nicely generated face for in-game portraits... but you still need a human to spot these things and then to crop/correct them out.