Famous AI Artist Says He's Losing Millions from People Stealing His Work
gizmodo.com
gizmodo.com
When the news first broke about it winning an award, I was like "well damn, I can see why—it's visually striking!" But the longer I look at it, and the more LLM-generated images I see in general, the less this one stands out to me. It's still pleasing, but seems less unique in both its strengths and flaws: the soft glowing lights, the tasteful muted palette, the random greebled details.
And it bothers me that the glowing orb/archway in the center has a distinct border around most of its circumference, but the part from 4 to 6 o'clock is fuzzier and blends into the nearby wall(?) :P
Kind of like a Rembrandt or something.
I guess there's a reason it's always been said "there's no accounting for taste" :)
Maybe it's not going to be too easy to reach the level of a "masterpiece" with only a non-owner to bring it into the world?
It doesn't pass a blind scientific placebo test.
This is the fundamental problem with AI -- if the government isn't willing to protect it, there is fundamentally no market for it. What does that say about the value of the actual AI tools? If the content and images you produce with them can no longer be protected, what value are the tools used to create them?
E.g. medieval craftsmen assumed that monopolies were good, but today we consider them bad.
If the food you produce with them can no longer be protected, what value are the tools used to create them?
If you accept that, does that mean source code is copyrightable but compiled code is not?
I'm kind-of ok with the raw output of diffusion models being public domain. It gets more complicated when The process is more than just Text->Image. Artists spending large amounts of time iterating, compositing, inpainting etc. are applying their knowledge of style, design, colour, and lighting.
They are nothing of the sort.
They are, at most, and even this is overly generous, someone commissioning a "fake artist" (AI) to make something for them
Yet “AI writers” isn’t really a thing journalists/bloggers refer to them as. It’s interesting from the outside, as neither a writer or an artist.
As a photographer you did not create the universe you are walking around in, you prompt your camera to look at a certain location and you push a button.
But he's wrong. If there were a fee to use this image, basically zero people would use it. They'd all go to midjourney and make their own better images for free.
> When an AI technology determines the expressive elements of its output, the generated material is not the product of human authorship.
> As a result, that material is not protected by copyright and must be disclaimed in a registration application.
but i think that's only if it's entirely prompt driven. you could argue that manual edits or additions would allow it to to have copyright?
For instance some people have always had more advanced editing programs than others, or different performance levels of their hardware to run the software which makes/allows the final product to be more effortless than the mainstream by far. Effectively with significantly different output from the same input, all unique no matter how you try to make all your PC's run the same, kind of like a video game where you get better resolution or framerate when you have more advanced hardware and/or software. Sometimes just a little bit better which could be considered incremental. Sometimes much more than that which can astound as a major leap.
When somebody comes along with a much more powerful hardware or software arrangement to output the same class of data, is it supposed to kick them out just because it's a major milestone that can not be ignored?
You could start with something like a completely original watercolor, and very well be using the exact same PC but naturally get two distinct printouts from the same digital file simply from using two different printers. With more significant differences than if one would have been touched by AI and the other one not.
Regardless of copyright, the person who submits the winning image to an image contest should really be the one to take home the prize.
There was a news report on it awhile back that was interesting to read.
One can dream, right?! This could cripple monopolies and bring more power back to small businesses and individuals, as the capitalistic playing-field is leveled.
The closest I've heard is a LLM generated book had portions of the book copyrighted by the "Author", but only the portions where the author had significantly altered the LLM generated text.
* AI companies claim that training should not be covered by copyright. This says nothing at all about the model outputs which can still violate copyright.
* Curation makes something a copyrightable work. This applies to photographers and machine generated art.
The copyright office's ruling doesn't change that second point although in practice it might raise the bar.
Are you sure about that?
I know one or two countries have "database rights" which give a special type of IP protection to collections of methodically collected facts, like digital maps and phone books. But I've never heard of "curation" being copyrightable in and of itself.
We live in an insane world where capitalism is a weapon which destroys and limits art, rather than art enriching us and the world.
Copyrightable output of pure AI is the nightmare situation. Imagine Disney trains an AI on purely content from in-house work-for-hire artists. They would then be free to generate an infinite amount of content that they claim ownership of.
Copyright law, like all laws are intended to benefit society, not individuals. They can be changed as circumstances change. Sure, lobbyists can have a strong influence and attempt to tilt the playing field towards some individuals, but the primary strength of lobbyists is that they show up. Change goes to those who show up.
So technically, he has zero right now and is begging the government to use force of law (i.e. threat of violence) to direct millions to him.
...I gave up on my first...
Or rather, it should get the same copyright consideration as all the artwork that went into building the models.
So still yes.
So as per your logic the answer would actually be "no, these images aren't exempt from copyright."
That said, were the images used by Adobe actually licensed for AI training, or was an existing overly broad license aimed at "showing this in an online gallery" interpreted by lawyers to include AI training? Because that latter option is being done by a lot of corporations (including Google) to make "ethical" AI models.
> Adobe does not train Firefly Gen AI models on customer content. Firefly generative AI models are trained on a dataset of licensed content, such as Adobe Stock, and public domain content where copyright has expired.
https://slate.com/technology/2024/06/adobe-terms-use-backlas...
> Nothing from the OP or the original AI Artist was about using Adobe.
The point is that even using basic photoshop tools is "AI art".
The magic wand is different IMO, because it's not creating any imagery.
Like code generated from models trained on AGPL/GPLv3 licensed code should also be AGPL/GPLv3.
Their work is copyrightable, and it can be argued the only "work" they did was pushing a shutter button and maybe some touch-ups. Nature did the rest.
We extend copyright protection to them; it's difficult for me to see why we wouldn't extend copyright protection to AI-synthesized works on similar grounds.
https://en.wikipedia.org/wiki/Monkey_selfie_copyright_disput...
Can we please end this delusional demand that property rights can be applied to data? Pretty please? If not now, then when? How much more unfeasible does copyright need to be before we move on? How much more unfeasible can it really get? When are we, as participants in a society, going to simply give up on playing this ridiculous game?
Copyright demands that we all participate in its own failure. That is, all of us but a select few who have already won the game.
Got it. So while substantial efforts have been made to claim that real artists—people who spent years of their lives working to produce actual works of art—have no legitimate claim to legal protection from AI companies, the people that should get legal protection are the people using Midjourney.
There are three types of labor-intensive AI art: (1) ComfyUI node graph editing, (2) canvas-based compositions that use AI as a filter or brush, and (3) large multi-day compositions typically built for film and gaming. Sometimes all of these manual and laborious methodologies are employed for one work.
We're so past the "prompting" phase: artists are building complicated workflows, manually drawing and sketching on diffusion canvases, and spending days precisely rendering the videos that they want. It's more like Photoshop combined with custom-built shader pipelines than Midjourney. Or a long virtual filming and editing process with deliberate intention.
You wouldn't call algorithmic manipulation by Photoshop non-copyrightable. If AI is just a "brush" employed by a human, and the human is doing a substantial amount of work, it shouldn't render the entire work uncopyrightable.
If you'd like to see what actual AI artists are doing, check these out to get started:
https://www.reddit.com/r/comfyui/
https://www.reddit.com/r/aivideo/
Here's something that was "prompted", but that took days of effort in pre-production, composing, and editing:
https://www.reddit.com/r/aivideo/comments/1ftvole/ouroboros_...
Here's something else that was built up from lots of node graphs and drawing and editing. This is a tremendously complicated work:
https://www.reddit.com/r/comfyui/comments/1fse25a/putting_th...
You can't possibly tell me that these are uncopyrightable.
And while you may or may not like these pieces yet, there is clearly something incredible is happening here. These are real artists. And as individual creators and small teams, they will soon be capable of taking on the likes of studios like Pixar without the millions of dollars of capital the big studios have employed.
Further, the official guidance was delayed again until later this month, and is currently open for comments: https://www.federalregister.gov/documents/2024/09/17/2024-21...
You can't have your cake and eat it.
I think the most compelling reference work right now is probably a collage.
> a piece of art made by sticking various different materials such as photographs and pieces of paper or fabric on to a backing.
And the law there is actually pretty nuanced and tangled: A collage, while not a compilation, may be a collective or a derivative work. A collective work has copyright protection, but a derivative work does not.
So basically, given the current environment I think we're honing in on a super clear and very helpful "sometimes"...
Adobe's models are 100% trained on works they own or license the copyright to.
Synthetic datasets are being built in Unreal Engine and with automated photo turn tables.
This argument is going to lose merit soon.
When midjourney and all other models built upon copyrighted work they had no license to disappears (and any models which were in turn trained on the output of Midjourney et.al.), perhaps.
> Adobe's models are 100% trained on works they own or license the copyright to.
I have to ask, was the license broadly interpreted to include AI training, or was that part of the original license? Because the former is what Google is doing right now with their video AI and scraping youtube.
But nothing force you to reveal the prompt you used assuming you are using the right tool for protecting your secrets.