The new forums will be named Coin Return (based on the most recent vote)! You can check on the status and timeline of the transition to the new forums here.
The Guiding Principles and New Rules document is now in effect.
AI Generated Images: Tech, Ethics, Impacts on the Art Community
Hexmage-PA
Registered User regular
Registered User regular
This is a thread about AI Generated Images, which is basically typing words into software trained on data sets of potentially billions of images taken from the Internet. I spent like two hours on the original OP for this thread, mostly talking about the potential dangers of this tech and how it could wreak havoc on human artists, but nobody here cares about that so feel free to post whatever you made in Craiyon or whatever.
If you are someone who would be interested in the original topic I have a counterpart thread in SE++.
If you are someone who would be interested in the original topic I have a counterpart thread in SE++.
ElJeffe on
+4
This discussion has been closed.
Posts
Battle.net: Fireflash#1425
Steam Friend code: 45386507
Usually automation at least is making machines do things that people don't want to do. People become artists because they want to be artists, and automating art devalues art and the process of its creation.
Further, these AI image generators are trained by feeding them images created by actual human artists without the original artist's knowledge or consent.
Plenty of people like doing things that are automated these days. People still knit, grow gardens, blacksmith, bake bread, build cars from scratch, etc because that's what they want to do. Some even make a living off of it. Their achievements aren't made lesser because I can get all of that for cheap.
If you measure the value of your creations by the monetary value they gain you then I'd contend you don't really want to be an artist, you want material rewards for being an artist.
That's influencing or interpreting, not stealing. The art was released to the public to be appreciated and now an AI is appreciating it, linking art to abstract concepts and creating new art. An artist can't finish a mural on a wall and then turn around say, "No robot eyes allowed. They're going to copy my style." It's a mural on a wall.
I agree that this has a high probability of hurting artists and designers, driving down the little regard they receive for their work and livelihood. But it also looks fun and can jump start the iterative process nicely.
* cue Das Racist’s “Taco Bell/Pizza Hut”
The last paragraph is the same bullshit used to justify paying bad wagess in fields as diverse as artists teachers and nurses.
If we cut your pay by 90% would you stay at your job?
The fact remains that the problem in your first post hasn't been solved. So for the people being automated it's still a problem.
Also tbh I'm not sure how these generators don't run afoul of copyright. Derivative works are a thing, and dumping an artists output into a machine that copies and remixes it is exactly that. You can see it very clearly when the generator is fed only a single artist's work.
Actual human artists are influenced by things they appreciate, and though an AI cannot actually appreciate anything it is still influenced by art it is exposed to, like a human artist. Unlike human artists, though, it could be possible to actually go in and look at what an AI referenced in creating an image, something that corporations like Disney may use to try and argue that their intellectual property has been infringed upon if there is record that it was used as reference (smaller creators obviously won't get to protect their own work in this way). I'm honestly kind of hoping that such corporations will end up getting litigation happy against the owners of AI art generators.
Human artists are also trained by feeding them images created by actual human artists without the original artist's knowledge or consent.
With an AI art generator I can type in "cute cartoon mascot character" and get this:
Who made it? Some code. What were its influences? Things it was fed and has been programmed to consult if a person makes that kind of prompt. What made it want to end up with this design, and is it happy with it? It can't want or be happy with anything; it's a soulless philosophical zombie.
It's missing the human element. A human can tell you what their influences are, why they made the choices they did, if they're happy with what they've made, what messages they wanted to convey, etc.
Honestly if there's anything that AI art generators might actually be good for it's to create images of surreal horror that a human mind could never come up with. I've seen some instances of AI "art" when looking on my phone at an art blog I follow that have triggered such revulsion in me that I would carefully avoid touching the image when scrolling past, I guess because I was subconsciously afraid that touching it would feel as unpleasant as looking at it.
So, I can sympathise with Dall-E at the very least
Celeste [Switch] - She'll be wrestling with inner demons when she comes...
Final Fantasy XII: The Zodiac Age [Switch] - Sit down and watch our game play itself
It’s impressive, but obviously not a replacement for specific artistic needs. Basically it feels like these AI bots are great at making vague images of 1970s sci-fi and fantasy book covers, but not much else. At least, not yet. It doesn’t understand the relationship between art and emotion.
As impressive as these things are, they'll most probably be relegated to replacing some of the more tedious work - the same way most CGI is used to fill in backgrounds details rather than being constantly front and centre.
And when it is front and centre, there's gonna have to be a whole bunch of people babysitting the output so it looks like it should. The same way that mascot from upthread would need someone to take it and make it less fucking horrifying before you could even think of using it.
Celeste [Switch] - She'll be wrestling with inner demons when she comes...
Final Fantasy XII: The Zodiac Age [Switch] - Sit down and watch our game play itself
Your first sentence is used to justify the status quo rather than actually improving things. The actual problem isn't automation it's refusing to change society to accommodate the fact that individual work is less and less needed.
And I'm a translator. My once impossible to automate job's going away at breakneck speed. If society leaves me in a gutter the problem wasn't translation software.
Some humans can tell you that. The vast, vast majority of art across history is anonymous. It doesn't become less valuable because I can't ask the creator why they made the choices they did. Great works aren't valuable only after talking to the artist.
I agree with you but at the same time as someone whose not a giant art fan that spends all their time in galleries I just..don't care? And I feel like that would be true of the majority of the population
They probably do. It's just never been tested in court.
Also, if you are an artist, then it might also be hard to prove that your copyrighted work was used as training data if prompts like "in the style of <artist>" are disabled.
Prompt - Dark cloaked figures walk across a scorched desert. In the distance the burning sun blazes down from between wispy clouds which provide the false promise of rain. Bold colors. Detailed.
Prompt - Woodcut of a cartoon duck fleeing across the sky pursued by a flock of starlings. Abstract painting. Black and White
Are these the best images in the world? no, they aren't. Do I feel something when I look at them? Absolutely.
And honestly? I find this uttterly astounding. The fact that a computer can do anything like this is just mind boggling, and Dall-E 2 is vastly better than Midjourney. And both are 10x better than anything which existed a few years ago. Whether it will be good or bad for society, I don't know. But I think this says something about where we have arrived, and that place is somewhere that computers can make interesting art. Art that I quite like.
Edit: And useful! This is something I can use on its own or other projects I work on.
No, it isn't. "You should be doing it for the love of it" is always an excuse not to pay. Nothing more. Would you do it what you do for free?
Also here's the thing. These generators only function because they're fed training data. And feeding them their own output to try and generate new styles is going to result in the same kind of garbled weirdness of feeding an AI translator back on itself.
So when it needs to be improved, where does the data come from? Probably the same place it came now, feeding in human artists. Who still won't be getting paid.
I never said people should do it for the love of it and not be paid. I said the problem is tying people's well being to what they produce.
As a reminder, this particular tangent started with:
Which is a farce. Lots of artists don't like what they do but it pays well. I propose the solution is ensuring everyone's wellbeing is cared for regardless of what they produce which, for some reason, you've interpreted as not valuing the things people create for some reason.
Yes, and? Where does that say artists shouldn't be paid or valued? When I enjoy art it isn't because of the price tag attached.
The problem is not "Not enough free art and images" when it comes to training these things. Deviantart (for all its sins) contains millions of pieces of art of all styles, produced overwhelmingly for free, and shared for free, for no more reason than the artist enjoyed making them. Over the last 3 years, I've spent a lot of time trying to learn to draw, and I've looked at thousands of different pictures for inspiration and to learn how they were made. I've also posted up dozens of things I've drawn for comment and sharing.
There are clearly better artists than me. However, there are thousands of free, public pieces which form an immense database of 'good' art to learn from. Here you go, hundreds of thousands of images, sketches and paintings, the entire collection of the stadel museum online.
https://sammlung.staedelmuseum.de/en/person?gclid=Cj0KCQjwxveXBhDDARIsAI0Q0x3-mndZs1WRrWeVIf34aYWEZkKOdTCgRA-cmy5XyRRjBHmKhWrXJrIaAtVxEALw_wcB
Noone needs to steal works to train these AI's. People have been making amazing drawings for hundreds of years.
This doesn't seem much different than human artists basing their styles on older artists and combining, which pretty much all do and the original artists aren't paid
I know, I was trying to express to my wife what a big deal this is yesterday. She was like, Oh, technology is slowing down, we'll never see changes like the emergence of the internet again. And I was trying to discuss with her that computers have just learned how to create art, and suddenly I can see how this whole neural learning network thing is going to change the world. It teaches me how I think. It makes me doubt about how big the boundary between me and the device I'm typing on is.
Thats my point here a bit, unlike with say, a neural network learning how to fit graphs or drive cars. The way the AI here seems to have learned how to draw seems a lot like how my son learned to draw, or how I've tried to get better at drawing. This is the most humanlike thing I've ever seen AI do. It's just astounding.
As a translator I also think my skills becoming redundant is great in the grand scheme. Translation software developing into something more accurate and accessible to the masses is a good thing. Same as everything else that takes time and effort to do on one's own.
I also have a safety net that'll ensure my well being as demand for my skills continues to dwindle. So I'll keep spending time studying and talking about language. Which is a way better situation than lamenting people's access to the art of communication IMO.
Artists probably want do art for the sake of self-expression. And the inherent rewards of the creative process. (they want to get paid too, but they could be doing any sort of labor and getting paid).
Making that 100th corporate logo or article art, in a manner that can 100% be done by an AI...
This isn't David or the Ceiling of the Sistine Chapel or stuff that their soul is burning to express.
It has value, and I respect the time that goes into it, but I doubt there's many artists that are driven by creating the sort of art that company is going to have an AI create instead of paying an artist.
I'm also a translator and agree (although I sadly don't have the safety net that you do).
I think that, as with machine translation, AI image generation is going to open up a lot of opportunities even as it kneecaps a particular profession.
News articles are definitely going to start using AI art instead of paying artists. Why not? Half the stuff in the New Yorker looks like an AI extruded it anyway, and it's wildly cheaper. A noble stance about paying artists won't last in the face of speed, ease and affordability.
A DnD chatter is already using two AIs to generate portraits for his NPCs and settings in D&D, and then refine them to fix anatomy and etc. Not only is the subscription cheaper than commissioning art, especially in that volume, it's faster and easier to tweak. You can easily illustrate your whole roster of characters.
Concept artists are going to have to fight for ground here, too. You'll definitely need skilled concept artists to design elaborate levels and characters for triple -A games, but the brainstorming, the pitches, the raw idea material, that can easily come from AI. Indie games are surely already using AI for portraits and background images. Concept art, like translation, will likely bend toward refinement and adjustment of machine-produced material.
My translation job, years ago, had a client that was almost entirely machine-translated, and for that our job was mostly revision. We got paid much less per word, but we churned through many more words. Some of us made more money on the automated material, in fact. But the quality of the translation dropped alarmingly. The client didn't particularly care, since they just needed the text to function roughly as it did in the original language. But a slow quality apocalypse is a plausible result of the incentives involved in automation, because making a decent product in 10 seconds is so wildly useful compared to an excellent product in five or ten business days.
Here's my favorite so far, "battle between cats and ferrets at night in the style of van Gogh "
I’m a content manager at a marketing department, and have been doing 3D and video work among other things for my whole career.
When we started using machine translation for texts the quality drop was immense. Just like EM mentioned. We ended up still using manual translation for marketing texts since not only does the “spirit” of the text matter more than exact translation, the packaging matters more. And with manual translation we can work with the same translators over time and they learn our expressions and tone etc. We did keep using machine-with-review for manuals though - the word-for-word translation matters more and the costs for translating a 50 page manual to 12 languages is astronomical to do manually. This directly opened up other markets to us because we could translate to Country X (local language manual legally required) to sell one production line and make a profit.
These AI tools will start being most helpful where intent and exactness aren’t premiered. The editorial sketch is a good example, it serves the function of a design element - it’s graphical content in essence. That specific job, editorial image or stock photo selection seems quick to go through a bit of a shuffle due to this. Like when we remade our website and due to a late design decision needed 300+ stock images. I spent over a month collecting the closest “okay” matches for the various industries I needed to depict. With an AI tool no doubt that could have been automated, either for image selection or generation of new sketches. I’d have been thrilled to spend my time doing other things and we wouldn’t have paid to bring in anyone else to do it due to the timing.
Similarly - we have something like 3000 products and they all have one product image on the website. All new products are rendered from a specific angle, they need not look photorealistic - just show the shape for help with customers selecting them. We still have at least a thousand of these images that are old photos from before 2008 that need to be updated but never will. There is never budget to do this manually, hasn’t been for the 9 years I’ve been at the company. If I could just feed the CAD files to a network and direct the generation for good results this could finally be done. I could even hire a consultant to direct the material selection and quality approval - creating a temporary human job where there was none.
These systems will not replace the need for humans in many other tasks. Even in the most extreme future case where an AI like this could take CAD files from an unreleased product and generate the equivalent of marketing beauty shots for it - there would still be the need for an artist intermediary there. Product owners in industry are not up to the job of selecting images for marketing. That’s just one example.
Concept Art is a highly specific skill set despite how loose it may look. Bulk concepts or mood boards can no doubt be done extremely effectively by these tools but when you start to get into specifics like actual design of a spaceship for instance - the concept artist will draw a 2D image that has a wide enough perspective that a 3D artist can map the perspective with guides and create an 3D model with just one angle of concept image as base.
So I don’t feel like the technology and tools will kill artistry in that sense. It is too wide a field to be a useful descriptor. It will kill specific tasks and jobs in the sense that Photoshop and FrameMaker did. People will shift to new focuses. No doubt these types of artistic jobs will change over time but it is over time and not tomorrow. A job like concept artist will still be a job in five years - but perhaps applicants will have “Midjourney 4/5” next to their “Photoshop 5/5”. It’s a tool after all, tools change. I think we’re more sensitive to the prospect of automation in these types of field because it’s more common to our generations and we lean higher education/creative on these boards. We’ve all been reaping the benefit of automation straight up killing manual labor for decades, I don’t believe what AI does for this field will be even remotely close to that.
Personally it seems highly alarmist to bring down the image from the Atlantic piece. I guess it’s fine to do so until conversation and sentiment around this matures a bit. But we seldom refuse new technology because it changes the status quo. I feel very sure this is a symptom of this being a white collar type of job which this has higher respect on the face of it.