You’re using two very similar cases in a very narrow slice of what copyright law protects as an example, when in reality copyright protections are much more broad and nuanced.
Disney wouldn’t have spent millions lobbying Congress or buying off Sonny Bono if they didn’t think copyrights were worth protecting.
graphic design can’t be obsolete, but graphic designers can become so. my career started well before computers became the viable tool for the job. and i have seen many people who did not have the training or experience sit down at the computer and bang stuff out. were the results outstanding? not really. but they were serviceable. and many clients (especially budget-conscious ones) will take serviceable if its at the right price. i now have clients who send me completely mocked up layouts and just ask for it to be tightened up for production. is the design good? no. is it quick and cheap? yes. AI will certainly be able to do that.
sure totally legal and 100% of the profit goes to the data owners that is was trained on, right?
Because of the way the art is used to generate more ar specifically in a particular artist’s stylet, i think that it should only be legal if the artist are compensated for the use of their initial artwork for training. This would require ai developers to seek out permission from artists (or those responsible for the estate) to use said artwork. Some would likely be reluctant as the artwork can be used to create pornographic images or images counter to what the artists believes in.
You could easily argue that by allowing people to generate artwork in that artist’s style could result in lost commission revenue. It is worse if the program is subscription based, because the developer is profiting off of the original artists work. Another solution would be a form allowing the use of their art and promises of royalty revenue for each prompt generated using that artist’s name.
If this was implemented, most ai artists would suddenly be using artwork from 50+ years after the artists death. However, they could also turn to stock images or creative commons licenses, e.t.c.
It depends on how prompts are phrased as well. If the prompt does not specifically include a reference to an artist’s style, but the image rendered still includes a figure drawn in a similar style, can it be considered plagerism? Or is it arguable that by making the image available on the internet, as long as it does not directly reference that style, can it be used as “inspiration” for the ai program? Humans may look up images as inspiration and create artwork related to those images.
Another issue in question is the ai’s ability to render recognizable photo realistic images of specific individuals. They still struggle with limbs and some features, but that is steadily improving. I think that will be more difficult to legislate because a human arist is capable of creating artwork that includes renderings of actual individuals.
I’m an academic librarian. While the issues I see aren’t anywhere near this, I’d agree with your interpretation, especially on points two and four, amount of work used, and market impact.
I have a hard time seeing how this is any different from sampling, where the sample is processed in various ways to be repurposed.
As we well know, samples are now compensated. Maybe we’ll miss out on the Paul’s Boutique of illustration, but I doubt it.
Most of the examples above are of copyrighted, trademarked characters, and the ones that aren’t are taking great liberties from Disney, for better or worse.
Not really. The “techbro” take would involve some way to sell the model, not to train and distribute a model openly and freely as is being done here. Yes, it is complicated as to whether a “style” is actually something that can be protected (especially given that human artists copy each other’s styles all the time and that is generally accepted) but what we are seeing here is a conflict between two creative people. It’s just that some people don’t see creating AI models as art in itself.
The default Stable Diffusion model is known to have been trained on the LAION image database which includes Shutterstock previews. This isn’t a secret.
But Stable Diffusion is just a compression technique. It can get incredible compression ratios, but it just reproduces sampled sections of the original image.The fact that its decompression algorithm allows one to mix and match doesn’t make it something else.
Scary the way this technology is going. It’s even more incredible they selected such a generic-style artwork to train the database, maybe for that insipid Disney / Cartoon Network, inoffensive aesthetic.
Wouldn’t be a post about a woman asking for her work to be valued without someone signing up to say it shouldn’t, I guess. 
Women’s stupid bodies make dumb things that are generic but ai is made by smart men who can someday remove the need for women and bodies alike!
/S
An earlier generation of GANs believed that cats always appeared with text captions.
@JHbadger said this dataset was LAION. Clip front is one way to browse the image/caption pairs, without downloading the whole damn dataset.
Damn there are some poor quality captions.
This whole ai vs meat artist thing that is happening on the internet right now is kind of tragic.
Not to mention I think some people get so legitimately triggered by the word “woman” that they are already angry when they see that “a woman” anywhere did anything.
Reminds me of people who believe that electronic music isn’t “real” music…
I also noted that artist says that what’s bothering her is that they are using her name to tag the art and ai generated which might be used to make images of literally anything, including some of the more unsavory crap people on the internet like to churn out. No one wants their damned name on that. It’s a valid concern IMO.
I know I shouldn’t but I want to say to the people saying “well her art is stupid anyway” obviously it’s known/valued by some number of people because people like it enough to train an ai on it and then use her name to identify it because enough people
recognize the name and attribute a style to it.
The adaptor of the software has changed the name of it and included a disclaimer “Hollie is not affiliated with this.” in the README (which we all know nobody reads).
So, that’s alright then /S
He then went on to release another artist version (this time a more conventional comic book style), with this message
“Speaking of that, Hollie was not pleased. Among a few issues with it, she didn’t like her name being associated with it and that’s why the name’s been changed and her non affiliation been made clear.
Now i don’t know Daly but i’d guess he won’t be pleased either. If that bothers you, keep it in mind and don’t download or use it.”
Internet police thyself
BTW the thumbnails do not do justice to Hollie Mengert’s work – there is far more texture and detail than is apparent. And there are already Disney-based versions of the software, classic and modern.
