#literally online newspapers and magazines are already capping articles with ai-generated images where they used to pay artists/photographer

[lethalhoopla]

Hey so! I'm gonna take the above in good faith as best I can and respond genuinely, because I know a lot of people, especially artists, feel very strongly one way or the other about AI-generated images - making it all too easy to miss each other's core points. And I think a lot gets lost in translation between 'sides', as even though everyone's responding to one another the unstated 'question' they're trying to answer is actually different. Namely: people against the AI trend tend to be responding to "is AI bad for stealing art?" while people who are trying to argue AI isn't inherently bad are often responding to "can AI benefit artists?"

I think once people start feeling strongly on a topic, it's easily to accidentally fall into the eternal human problem (that's even easier on the internet) of "inventing a guy" - in this case, a guy that feels there's black and white morality here in either direction. Some people feel strongly enough that they think even in spite of the potential good that all AI generating attempts should be shut down, while others feel strongly that there's enough good to be had they think everyone should just try and make the best of it all.

It's not misinformation that the op said "CSP is including a feature for ai art so it's easier than ever to steal credit from artists" - I can't be sure of their intent, but hell, they even put the pic of the tweet itself where CSP said they weren't feeding Stable Diffusion user's art. They weren't hiding that fact. The stealing in question isn't necessarily about CSP users - but rather, that SD is built on stolen art. Anything SD generates is inherently built on incredibly shady and dubiously ethical practices of scraping art from the internet.

(Insert a thousand arguments here, I know. But it wasn't until SD 2.0 - released Nov 23 - that SD users were blocked from easily and explicitly generating images to mimic certain artists styles; say what you will about 'learning' and 'style mimicking' in flesh-and-blood artists, the fact that those artists will never have an option to fully remove their art 'data' fully once its been used is pretty fucked up.)

That said, I don't want to put words in people's mouths, so I'm going to focus on why people are pissed about CSP + Stable Diffusion.

See, plenty of us that aren't happy about this know that CSP isn't feeding user's art into Stable Diffusion! That's definitely a relief. The problem that a lot of us have - both with this 'collaboration' and with SD itself - is that Stable Diffusion is built on stolen art.

(And, to be clear: art includes photography. Stolen photography is also feeding these things, and that's just as messed up.)

In other words, no matter how good your intentions, even if you aren't making money off of - for example - using the 'AI' to generate a background - it's impossible to extricate a result that hasn't benefited from 'learning' from countless works from artists who did not consent to their work being used to train a machine.

Supposedly, future versions of Stable Diffusion will have ways for artists to opt-out of their work being used as training data. Still, that's a nebulous promise; and Stable Diffusion is open source. While I'm normally a fan of open source software, what that means is that people can and do simply make their own version and add back in the 'training sets' that they want their generated images to 'learn' from. In other words, it's extremely easy for people to just... add those opted-out artists right back in to their 'own' version of SD. (The discord for it is rife with advice on 'training' your own SD instance...)

There's a viewpoint I see from people who have less-to-no problem with programs like SD who say that well, other artists learn and get inspired from other artists too, and hell, working artists on creative teams often even get shown specific artists' work and are told to 'do something like that. This discussion gets even muddier than the rest fast, because frankly, it gets philosophical as hell - how is machine learning different from a person learning/training their art skills over time, etc etc? Seriously, dissertations can be written on that.

I think where I personally land is based more on the outcome of this machine learning.

Essentially: while in a perfect world this could just be a cool inspirational tool to boost artists, we live in a capitalist, corporate-profit-driven world that will use these tools in every worst way and leverage the technology to cut out the 'lower tier' artists on their teams. Entry-level artist positions, critical to even building up the experience and further skills necessary to make it to senior positions - or build up enough connections/portfolio to snag one of those spots. Will every art position be lost? Likely not - top tier artists in the field will still be kept to do the highest concept/proof work, or finesse whatever AI generators give, etc etc. But there's a great Kotaku article that interviews artists and creatives on all sides of the issue that puts it well, I think, and while interviewing one artist:

Jon Juárez, an artist who has worked with Square Enix and Microsoft, agrees that some companies and clients will only be too happy to make use of AI art. “Many authors see this as a great advantage, because this harvesting process offers the possibility of manipulating falsely copyright-free solutions immediately, otherwise they would take days to arrive at the same place, or simply would never arrive”, he says . “If a large company sees an image or an idea that can be useful to them, they just have to enter it into the system and obtain mimetic results in seconds, they will not need to pay the artist for that image. These platforms are washing machines of intellectual property.”

To be completely fair, this article also includes plenty of quotes from artists who have less of a problem with AI-generated art! And I understand where they're coming from. Some are just viewing it with mild resignation and 'well people steal my art anyways so what's the difference with a robot', and others as a cool potential tool for themselves. Some of the most optimistic - and I love their ideas, honestly, they genuinely helped me see the pros beyond the cons - are focusing on the ways they can broaden their own art horizons and grow even more as artists. I recommend reading that article for their PoV, too.

Honestly, I want to be excited about AI-training here. I love sci-fi shit. I love thinking about the future where we get robots who Are People.

'AI', however, is not artificial intelligence. Not even close. At best, virtual intelligence; and more blatantly, just a complicated machine with very well-tailored tag audits. What's undeniable in AI 'art' is that it is wholly built off of other people's time, sweat, and hard-earned skills. I could go on a tangent here about that whole difference-between-machine-and-people bit, how there's just inherently more effort in an artist that requires dedication and determination and ingenuity and developing your own eye and skills, but honestly, I'm mostly horrified by the lack of respect for artists, and the lack of choice.

I think one of the more notable examples of how this has really fucked over an artist that originally had a 'hey this'll be fine' stance is Greg Rutkowksi. He's a prolific and phenomenal artist, works a lot in concept art - and just so happens to be one of the top used 'prompts' in AI generators. I highly recommend giving this article about his case in particular a read.

Basically: by the time he realized AI-generated images with his work as tags were dominating search engine results rather than his own work when searching his own name, it was too late to try to get the AI art generators to stop using his work. Hell, they couldn't take it out if they tried.

Worse yet, even with efforts to reduce the ease with which prolific artist's styles can be copied, he's known as a 'shortcut' to a high-quality image generated.

Let's be clear: stealing art is stealing labor.

To feed an image-generating machine artworks that were not freely given is to feed someone else's labor into a machine they have no say, no gain, no return from. Artists spend years developing their skills. Years. Even if they're primarily inspired by one artist - big if - they are still developing their own style that is colored by their visual analysis, the quirks of their mind and taste, their experiences, hell even the fine motor control of their hands. No two artists are exactly alike - even if one's damn good at mimicking styles. That in itself is a skill to be valued.

Now, to be fair, as mentioned above: Stable Diffusion's latest release, 2.0, has severely curbed the ability to mimic exact artist's styles. And I do want to give credit for that- someone, somewhere in that team, listened to the anger and frustration of artists - or at least the ethical and possibly legal concerns. (Because yes, the generator used copyright protected artwork as much as 'art posted for fun'.) But to be just as clear: Stable Diffusion is open source, and the data needed to simply add unwilling artists' works back in is easy-access. There is a vocal and active community supporting this.

With all this in mind... I'm still against any art program collaborating with current major art generators. They have other reasons, and I get chasing the tech trends. I get why some people, artists included, are in favor of just using the tech now that it's there.

But I hope it can be understood why there is such a vocal opposition - and why it might come off as a 'moral black and white' matter; they're speaking strongly about it, often in short bursts. Pretty much no one's gonna read all this just because it's long - but it takes minimum this much to begin to address anything. And I left so much out of this. Point is, though: faced with everything as-is, it's hard for someone strongly against current models to see other options beyond starve out the current programs. Don't give them an inch. Do not support even tangentially with your money.

I'm not saying anyone above has to agree with that path. Just - give some more thought to where those of us against it are coming from.

I doubt I'm going to change anyone's mind. That's okay. I think a lot about this kind of stuff, as both a tech-minded nerd and an artist; and I think a lot about how so much is lost in translation when we're all inclined to respond in a handful of sentences. A lot is lost in translation.

I want there to be collaborative efforts to make art, and broaden what possibilities are out there! Honestly, I wish there were more efforts in this quickly expanding space to make opt-in-only, from scratch, AI generators. Hell, make one from public domain and free to use images! Give artists tools and resources to help - not push them out of their own damned field.

I'm not fool enough to think any of it's going away. But I'm also not fool enough to think that anything built on current generator models (Stable Diffusion, Dall-E's OpenAI, Midjourney...) can possibly be fully ethical. And that's the rub - that's what sucks about this new feature from CSP. It's built on inherently unethically sourced work. Labor. It's hard to pretend I'm anything but frustrated when the shoddy foundation of this whole AI-generated image business continues to be ignored.

... Also. Hey. Hey, everyone? Don't fucking pirate indie creator's projects. Toby Fox is one fucking guy making a game with only the help of a couple people for graphics and such. Don't steal shit from creators like that. Or from indie authors, indie comic creators, or indie musicians, or etc. What the fuck. That's not the same as pirating a movie from goddamn Gisney or fucking microsoft word or whatever. Being broke sucks, I get it, I've been there with fuckin $4 in my account before. I'm glad the above commenter bought it eventually, but for real. Please realize the difference between corporations and indie creators.

CSP is including a feature for ai art so it's easier than ever to steal credit from artists