Forums » Smalltalk » AI Art Generation: Beneficial or Problematic?

I mean, it's always important to bring the topic's up, since... well, we're kinda in it these days and it's critical to get an idea of what's going on it.

Regarding the topic of generations, you mentioned this:
To which, as you probably know, there's this: A1111 or this The one I use. There are already a lot of open source sources for anyone to generate AI locally (and unlikely before you don't need a very expensive rig to make these programs run... it helps, mind you, but ForgeUI can run on a 1000 Nvidia series GPU, if you give it time, so... almost anyone who has a PC can use them for free. I'm currently one of the "high end user" (for local), and I'm using a 3090 (which costed me 650€ used), but a couple of months before, I was forced to switch to my old, worn 2060 (300€ brand new, used is gonna be around 200€), and I was able to generate the same thing I do now, just slower. The cost has dropped dramatically these days, so it's not so "exclusive" anymore (in a mere term of financial expense... in terms of skills, takes a little bit of training)). I mean, if you are willing to use Linux, you can get a 7900XTX for 1200€ or less, and it's the second-best consumer GPU for AIs in the market; that's not the insane amount one would expect.

Now, online AI generators like NovelAI, Civitai, or others, if we take this into account, cannot be exactly considered just selling the generators themselves (since almost everyone can do it, 95% better than them, too, because you have much more room to customize the work). They are either selling
- A service (Midjourney is a good example, but also NovelAI: they give away expertise, fine-tuning, a prompt refining system, custom-made models, etc.)
- Computing power (Aka: GPUs, electricity, a place to store said Hardware, etc.)

I won't deny that the offer is bloating the market (which is to be expected: AI is the new buzzword, and everyone wants to try it... sucks that most of those generators are a pale imitation of the real, achievable high), but I can't be mad at them either: they found an opportunity and took it. Also, mind you, I'm not saying the free spread of AI art generation is gonna solve the shitty attitude corpos have onto their employees: I'm one of the very few jobs in the AAA industry who's outside this blood carousel, and I see every year people axed on the chopping block like they are more cattle than humans, and it's sickening (industry of dreams my ass). However, the point made was to consider that stigmatizing AI art generators (and users) will only empower said companies, not actually stop the system behind it at all.

I personally always discard this argument because it's well-known that thousands of Artists use copyrighted material to make their work, sometimes for commercial purposes (commissions, in a way also Patreon could be seen that way, but I'm of two minds about donation systems), but fair use laws protects them from most retaliations from copyright owners like DC Comics, Marvel, "Nintendo" (this one is... a tough nut), etc.
While it is true that AI generators scrape Internet for images in the likes of millions, it is also true that the work is transformative; carbon copies are unlikely, and are confined to the user's intentions (maybe using a trained LORA on that character). In this case, then, I view it as the exact same of someone making a fan comics of Batman having a relationship (maybe NSFW too because "welcome to the Internet") with the Joker and not having to worry for DC Comics coming for them.

I mean, this entire site exists because of this concept...

To clarify, your argument is basically "Because there has been a history of disreputable use of IP on the internet since the internet, I see no problem with continuing such disreputable methods going forward"?

Also, Disney and Nintendo are nortoriously litigious when it comes to their copyright. It's just a matter of whether it's worth the time, money and energy to go after whatever subject. Anne Rice was known for going after after all walks of fan artists/authors while she was alive as it was (unfortunately, maybe) well within her rights to do so.

No, my point is"Because there has been a history of a fair use of IP for their own personal gain, artists' point of Model generators\fineturners taking their work without permissions is hypocritical because you can't have the wife drunk and the barrel full.".
I do not find anything wrong with an artist taking Batman and making them hanky panky with anyone, so long as they add value to the established IP (yes, porn counts as value if the story\drawings\etc. appeals to others) and don't just make a carbon copy out of it. But then the same point stands for AI: artists complaining looks hypocritical to me because they are doing that now that the tables have turned, but before, they were the ones benefitting from it (and that was, and is, fine). The reason why Disney doesn't go after all the artists is precisely because of the fair use law: they could win a court case, but that would cost them too much money and public image to gain anything, and they could lose too. Without a fair use law like the one we have right now, they would always win, which would be awful.
Nintendo is a different case because of a cultural difference: in Japan, fair use laws do not exist, and they are willing to lose money to avoid their copyright being used without permission.

The same fair use laws also make the AI noncopyrightable, mind you, and again, that's 100% fine with me. If I'm benefitting from the fair use laws, it should also be enforced that I lose something out of it for the sake of being fair. With that argument, the artists are saying, IMO, "It's fine as long as it doesn't hurt me, only others".

I'm agreeing with you, actually, on this point. I'm not a big fan of people making an income as an artist exclusively creating fanart either. Doing so actually does violate Fair Use laws because the purpose becomes primarily a commercial benefit for said artist. I'd also say having explicit pornography of certain characters could be argued as devaluing the value of the original IP, especially if it's IP targeted at children.

But that's also literally what you just dismissed within the video Sanne posted as valid argument and what has been at the core of every argument in this thread. No one cares if it's used for personal use so much as it's use in commercial and corporate ventures without permission.

How can AI work be deemed transformative when there are no legal definitions or laws in place for AI artwork, as you acknowledged already? What defines transformative? Because AI can and does create perfect copies of existing artwork. What makes that transformative?

I wrote this comment on the YT video in response to someone saying "Well what's the difference between AI learning art from copying existing work and a human being learning art from copying existing work?" and I think it directly counters the argument you are making about scraping existing images:


Edit: I also want to point out that saying "I discard this argument because..." is deeply annoying to hear. You can disagree with an argument, but you don't get to discard a valid argument because you disagree with it. That's a very manipulative and exhausting way of having a discussion about something because you're saying it doesn't matter what someone else is saying, you're not even considering it. That's not a good faith discussion. Counter an argument with another valid argument, don't say you're discarding/dismissing it, please?

I'm going to point to an earlier comment I made:

I'm doing this because your responses have made it clear that -- and I mean this without any ill will -- you don't understand copyright and fair use either in general or as applied to this situation.

Generative AI algorithms don't just look at images. The companies creating them make copies of works, in full, to add to training databases. Now copyright protects owners of intellectual property when it comes to copying and reproduction of their work for pretty much any reason. You need a license to reproduce someone's work no matter what you intend to do with it. This is very different from artists creating fan works.

Commercial intent is much less important to deciding fair use than potential commercial damage to the holder of the IP. When people draw fan art, the company that owns the IP is at minimal risk of being materially hurt by it even if that person chooses to sell their art.

The same is not true when it comes to individual artists these tools threaten to replace in the marketplace. That active threat to livelihood is a considerable weight in terms of deciding fair use.

In terms of court cases, most have been up in the air and await judgements. Some claims of infringement have been dismissed, specifically in a case where plaintiffs attempted to argue that every work produced by a model was a direct derivative work. In that case, however, the original unauthorized copying was not decided on and still remains to be seen.

Here's a summary from https://copyrightalliance.org/ai-copyright-courts/

In short, I don't think you should be dismissing arguments offhand because you misunderstand both the intention and application of fair use. Copyright and fair use are also woefully out of date laws that prioritize the interests of corporations much more than individual creators. Ethically, the theft engaged in by the tech companies developing generative AI is quite clear, and artists benefitting from fair use or individually engaging in dubious behavior has no relation to the discussion at hand.

This is also speculative but I want to add that there is a difference in taking a single instance of an image from the internet and altering it (digitally or otherwise), and taking an image from the internet and storing it on a database to be distributed a theoretical infinite number of times.

The difference is that you have one single instance of something that is stored on your local PC or is a piece of paper in your house, which is (generally) private or accessed by a small amount of people. An image on a database is publicly distributed thousands of times, and then altered client-side to fit the client's needs. It doesn't really matter that the end-user is using something that might count as fair use when the curator of the data they're using is fullstop distributing an unaltered version of the work. They have the infrastructure to redistribute other peoples' works, and that's what's likely to be brought up in court.

Again, this is speculative and anecdotal but it wouldn't be the first time the technology is specifically targeted. Pirate sites get hammered this exact way all the time, even though making your own backups of material you own is protected in the US. YOU can have a digital backup of all your anime DVDs, even if you get them from a torrent site, but the nature of torrenting means that you're distributing while you're downloading and that's how people end up getting arrested for it. It's the technicality behind "you can make your own backups but not download them." You absolutely can just download your own backups from your own private server, it's when that server becomes a distribution vector that you land in trouble with the law. (Historically speaking, anyway. Civil law is actually very wonky and you can't assume much with it reliably.)

That is definitely an interesting bucket of technicalities to add! The whole thing is very complex.

I do find it interesting that many of the companies that would aggressively pursue piracy on the basis that distributing a copy of something is directly financially damaging, are now gleefully experimenting with generative AI, a technology that runs on unauthorized copies.

AI generated art uses art taken without an artist's consent and I cannot bring myself to ever use it. If there was a model where the dataset was ethically acquired, sure I might make some use of it for landscapes and whatnot, but I would rather commission an artist if it involves a character.

If i ever wanted to make an OC, then, i’ll request an artist for that.

I’m never inspired to make such character, but a real person is more legit than a machine itself.

Not to mention that you got so many cases of art-theft online..

A theft done by a person is easier to track than a machine, so for the AI, i cannot be always aware if something have been stolen.

I wouldn’t consider AI as problematic if it didn’t tried to invade so many things, voice-acting included.

I have to go with both because, even though AI art generation is bad, it has its good moments. You can use AI art for concept art and inspiration but absolutely not for profit. Otherwise, my view on it is neutral, but I'm leaning toward the beneficial side of things.