AI-generated content: Legality, ethics, and the nature of art

Figured that since it's the new big thing sweeping the globe, and how it'll likely have a profound impact on media going forward, I decided to open up a thread specifically to discuss the topic of AI art generators like Dall-E, Midjourney, and the like.

It's a polarizing subject, but I think there's merit in debating the various issues surrounding it.

Such issues include "is it ethical for AI art generators to be trained on data scraped together from copyrighted works, and how is it different from humans getting inspiration from art?" and "what do you see as the future of commercial artistic endeavors going forward (comics, game asset creation, animation, etc)?"

EDIT: Expanding the topic to all forms of AI-generated art, including creative writing and music composition.

Edited by danime91 on Oct 11th 2022 at 12:12:24 PM

In my opinion, AI art is sort of like tracing.

It's okay to do for personal usage, maybe training, but professionally? Straight up art theft.

I think it's also effectively useless for anything requiring actual creativity. I'm writing a sci-fi novel. I can't use AI art to do concept art for me because it's just mushing existing things together and it's not going to understand what some of my details actually mean, and I don't think this is ever going to be overcome.

The other element is just that there's a contingent of people who are pro-AI art and who are just being the most colossal jerks about it. And no, I don't think making AI art actually makes someone an "artist". People are inspired by other artworks all the time, but AI art literally remixes bits and pieces from other artworks and can't really come up with anything genuinely new.

I think it's a neat intersection of language parsing and content generation, although I do wonder if this is too niche a thread. We've got a computer thread, then an even more specific AI thread, and now a thread specifically for using AI to make art?

Sure, but think of all the entities that demand art or imagery without demanding any sort of context or meaning. Look at all the bland corporate art out there, just meant to fill space. There are plenty of applications for AI art, a lot of which doesn't look half bad, and I expect it will disrupt the field a great deal in the coming decade.

Edited by BrainSewage on Oct 6th 2022 at 11:39:20 AM

And there is a contingent of artist who "Inspiration" is just stright out copy what other people do, in this many will not care if AI do the same.

@Raineh Daze Yeah I considered just using the AI thread for this, but this intersects with so many other topics, like the philosophical meaning of art and creation, artists' rights and their less-than-stellar treatment in the workforce, and works that use AI-generated art and the merits or lack of merits thereof, that I figured it would probably end up going off-topic a lot in the general AI thread.

Personally...

• I am of the opinion that AI would not completely obsolete artists. At least not until the point where Ridiculously Human Robots exist.
• I do not consider people who just put in prompts for the AI as artists, but lazy posers at best, as elitist as it sound at a glance. However I do think it's fine to use AI generated content as either inspiration or as part of a lager piece (for example, the basis of a texture for a given object), as long as there is still human input involved.
• I also do not think it will catch on with the general public as long as it is associated with get Rich quick pyramid schemes (like the NFT)
• Regarding the mass produced filler art industry. I do think there will a massive disruption there, as the Law of conservation of detail is a doubledged blade. However I also think there will still be people hiring actual artists for specific pieces for that same reason, as well as people not using AI Generated art out of principle or quality related reasons.
• Memes are fine
• The nature of learning algorithms means there is some element of plagiarism involved in such, so caveat emptor.

As some asides I found out about how that intersected with the anime sphere:

• First there is an anime oriented attempt that ran into some legal issues ( apparently the law actually caught up to lolicons this time) on top of using Danbooru and similar sights for it's references (which amplifies the plagerism issues)
• Secondly. Hololive talents Kiara and Mori banned the submission of AI art on their hashtags

I do thing it will catch up with the public simply because it will generate art at cheaper price and the more contect the AI get the "better" will be and consumer have a tendecy to ignore how the sauce is made as long it get made.

It has an uphill battle due to it being push partly by techbros and crypto-grifters, so it might still wind up with frosty reception on more extremely online spaces.

By the way - i currently have the Novel AI's highest subscription level, which translates to essentially unlimited image gens for anything other than the "Img 2 Img" tech. If anyone wants to test the limits and capabilities of it, feel free to make a request in here and i'll answer ASAP.

I'll also post my own tests around here, so you guys can have an idea of what it can and cannot do (amongst the later - it can't do stripes using a *specific* color in a piece of clothing. Believe me, i tried).

It wasn't so much as "legal issues" as some Twitter users saying it *should* have a lawsuit brought against it. Ironically enough, it was received warmly in the Japanese community.

Edited by Lalapolpolpol on Oct 6th 2022 at 5:47:44 AM

I do think Cover of novel will a place for AI art since many author often go for a feeling or editor tell them what to add. So they will probably go for a cheap AI art and call it a day.

@llppp Fine I'll bite.

"people wringing their hands over the future of art"

The reason for this choice of words is because I saw it used for dall-e on another site for a shitpost and the results there were off the mark (the closest interpretation to them were people worshiping analog mediums like paintings), so I want to see if other AI gens can get a closer interpretation of that.

Moving on. I have to ask, how do you think museums will react to the existence of AI art? in general or in the case of specific pieces like "Loab"?

I do think it would be amusing if, due to so many people generating a ton of mediocre AI art, the art generators start succumbing to inbreeding in a sense and just get trained on massive amounts of AI art and thus churn out more "AI art-like" results. Because as they are now, AI-generated art is pretty easy to pick out due to them all sharing some common quirks.

Is there actually a handy list of these things and how accessible they are to use? I know they range from 'web platform' to propietary, with 'freely released but you'd better like setting command lines up' somewhere in there.

Here's one made by me:

Dall-e Mini: The Firestarter, Probably the most used, at least until some time ago. Very bad image quality but is surprisingly witty and smart when interpreting tags and prompts. Free.

Dall-e 2: The decaying champion, The most famous, but i personally find it to be the worst one. Is good at following prompts and such, but suffers from a lack of configuration options and pics suffering from "stock image sindrome". Requires an invitation, gets 15 gens free montly, can pay for more for a stiff price.

Midjourney: A growing star, know for it's surreal, dreamlike feel to it's images, though similar to dall-e 2 it lacks configuration, but at least the images look *cool* and aren't "stock image derivates". Has recently adapted code from Stable Diffusion and is probably the best general-use generator for those incapable of running Stable Diffusion in their own PC and/or through Google Collab. Paid, montly subscription, them unlimited use.

Stable Diffusion: The Emsemble Darkhorse, was published this August and since them has grown into arguably the best free generator until now, if not the best one all-around. Can generate everything with great quality and is *extremely* adaptable thanks to it's source code being open for everyone, with a dev team actively working with the community and listening to feedback. Free, though it requires a *strong* PC to run locally. Can be used both on Google Collab and a official website if you can't do that, though the latter option is pay-per-gen with a internal currency.

Novel Diffusion: The New Prodigy, developed by the same disitents of the AI Dungeon incident who made the Novel AI text generator. Is limited to furry/anime art as of now (though you would be surprised at how far these two categories can go), but it's undoubtedly the highest quality generator so far, creating art on a professional level constantly and with little imput on the part of the user. Also uses a alright different "tag based" prompt style that can throw off AI veterans off the track. Paid, montly subscription, them unlimited use.

Edited by Lalapolpolpol on Oct 6th 2022 at 8:19:39 AM

Hmmm, there was a press release about something based on stable diffusion a month or so ago, I forgot the name though.

(I have a 3090, for running generation I should be fine locally)

Something I've noticed browsing deviantart (aside from just the sheer number of accounts less than a month old spamming dozens of AI-generated pieces) is that if you go for anime style they all look like they were made by the same artist (who happens to have a lot of difficulty with hands so they hide them in various ways).

Edited by danime91 on Oct 7th 2022 at 7:55:43 AM

Judging by images on wikipedia page of Stable Diffusion, it seems it can mimic different anime artists (at least when drawing Reimu..)).

AI generated art is hopefully a fad that will lose its novelty eventually once enough people realise its unprofitable

Yeah, it's uh, really difficult to actually make money off of. There's also the really thorny copyright question which is only going to get worse over time.

The people going on about how it's the wave of the future and traditional artists are obsolete are, tellingly, the exact same people who constantly stole art to mint as NF Ts and who won't shut the fuck up about how their completely deranged understanding of how things work is supposedly accurate.

Doesn't help that that particular crowd appears to think the height of art is vaguely attractive women with huge breasts and a lazy eye.

Edited by Zendervai on Oct 7th 2022 at 11:47:25 AM

Beyond the "is it art? is it a fad?" discussion, I think the legal aspect needs to be pressed. I've seen lots of threads showing how some of those can be trained to mimic specific artists styles. The AI is capable of that because of all of their work being fed into- meaning services are using their art and charging other people for it.

But then you run into that de facto just being "someone looked at all your art and copied your style".

I think there's a difference between copying a general style and literally blending bits and pieces of different works together.

Note that tracing isn't considered acceptable.

And have fun defining that line in regard to normal art, too.

The problem is that these don't just blend art together. It's all pattern matching. Direct copying isn't involved.