This may not be the reason he's "losing" money here, but I was struck by how much less... impressive that piece is to me now compared to when I first saw it. (Visually impressive; I've always known it was from Midjourney.)
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
AI-generated images are interesting in that, in general, the more you look at them, and in particular the more you look at the detail, the worse they look. This is, arguably, one simple way to distinguish them from actual art.
Pretty images are worthless, there's no point staring at images that have were made without intention or effort. Making a prompt is hardly any effort and more of a lucky hit.
Look for ComfyUI tutorials on YouTube. Creating something unique is elaborate and requires developing skills no different than Photoshop or any other digital art tool.
Based on his argument, the human component should and rightfully be copyrightable. In this case, it's the prompt or conversation he had with the AI. That should definitely be able to be copyrightable, but I agree that the image itself should not.
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?
>in this case, it's the prompt or conversation he had with the AI. That should definitely be able to be copyrightable, but I agree that the image itself should not.
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.
Source code for software is pretty exacting - the compiler or interpreter follows discernible rules to reach an end product. An image prompt has a random element to the end result - no one is really able to nail down the rules to describe exactly how the end result is produced. There's not so much creativity, just very abstruse statistical calculation.
That's a bad analogy. Food is consumed exactly once and has to be produced again each time, and thats where the majority of the value is, even for food that is produced many times over from a recipe.
Dear journalists/others: for the love of all that is good, please for f's sake stop calling these people "artists". It reinforces the lie they tell themselves.
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
I always wondered why there was an immediate rush to call them AI artists despite the first and predominant use being writing. Even a prompt is just writing text. Even with all the fancy visual tools, you still need to write a prompt somewhere along the lines.
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.
This ^^^. So many cannot see the blatant parallel with photography here -- there is a rich art historic debate regarding the value of photography that went through very similar stages of rejection (and ultimate acceptance). This is a fantastic example of George Santayana's famous aphorism.
This artist won a $300 prize at the Colorado State Fair Fine Arts Competition 2022. I'm a bit confused what millions he's talking about but in my opinion the best bet for him would've been to commission some artists to create a real painting of his work and strike the iron while hot. Did anybody think that gen AI artwork would be a hot market for long?
I assume he added up the number of times people have used the image and multiply it by some fee.
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.
According to my communications with the US copyright office, you never get a global copyright by editing. At best you get a copyright on your changes - and that only if they are copywritable on their own. Minor changes to wording or tweaking an image would generally not pass that threshold. It is a case by case judgment as to the status of any individual change.
This is backed up by how they have granted copyright to an AI generated and human edited book. The human only gets to get copyright on the parts they "significantly altered".
There was a news report on it awhile back that was interesting to read.
Also the other way - if you use an AI tool to edit an original image, is the output copyrightable? Is the answer different if the AI thing is doing a generative fill vs a bigger change? I guess this is the more financially important version of when Facebook was tagging any use of AI-powered tools in eg Photoshop as an AI generated image.
Yep agreed. My understanding is that Adobe add some sort of metadata if you use those AI powered tools, and Facebook were displaying AI generated image if you uploaded an image with that metadata present. I guess I'm just curious whether the output of those tools can be copyrighted. Is it a boolean like Facebook treated it, where as soon as you use any of these tools it can't be copyrighted, or is it more complicated?
From a more abstract viewpoint, what's the difference between any computer-aided editing at all versus none whatsoever?
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.
It's fun to ponder how modern society will evolve if this same mindset is applied to other domains that generative AI is eventually used for, specifically sciences. If AI (art) works cannot be copyrighted, there's hope that the same would apply to the other domains and prohibiting AI-generated works from being patented. Imagine AI finding 50 new ways of making insulin that don't violate the existing patents...
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.
Last I heard, nothing generated by an LLM can be copyrighted. Copyright is for protecting the intellectual work of humans, not computers.
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.
The irony is so thick that you can eat it with spoon. Is like stealing meat from a market and then arguing that it is his, because he already transformed it into a meat pie, and that's his own product now. :)
Y'all really want to be mad at this but this is one of those tragic cases where the worst person is correct.
* 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.
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.
Genuinely asking: would/should the ownership rest solely with the person promoting the model, or would part of all of it belong to the model itself (and/or its creator)? My hunch is that this is a somewhat unprecedented scenario, so comparing Midjourney to a tool like Photoshop wouldn't be completely accurate—in the sense that of course Adobe doesn't hold the rights to things you make in Photoshop—but there could very well be established cases I'm unaware of!
No, curation of a public domain painting does not make your scan of the painting copyrighted, even is someone somehow take your scan in his own book. I think it's called integration works, and basically the copyright remain with who owned them in the first place (cc/public domain images makes genAI a very fun case for IP lawyers. I think the 'nah, not human=not IP' is a good punt).
I don't think copyright applies to the image itself. The work of curation can be copyrighted, but here it's trivial, just selecting one image. If he published a book of AI-generated artwork, the curation of images could be copyrighted as a whole, but each individual image would not be copyrighted.
What happens if he doesn't publish the book during his lifetime, and later his estate decides to release it but no matter how hard people try, nobody can figure out whether or not any AI was employed?
The AI generated content would not be copyrightable. Manipulations of that AI content by people would still be copyrightable.
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.
It means that studios will shift how they lobby for copyright law as they train internal AI models on their asset catalogs. Eventually, Allen's claim will be mainstream.
Disney may well buy AI companies that may compete with them (like a next level acquisition equivalent of when they bought Pixar) and copyright everything instead.
We live in an insane world where capitalism is a weapon which destroys and limits art, rather than art enriching us and the world.
The inpainting tool in Photoshop is AI. If you use the inpainting tool, congratulations you're an AI artist. Should copyrighted of any images produced with the inpainting tool be ignored?
But the images that went into building the models are copyrightable. Adobe has a license to use them for AI training, but the authors still have copyrighht.
So as per your logic the answer would actually be "no, these images aren't exempt from copyright."
Nothing from the OP or the original AI Artist was about using Adobe.
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 trains its models on Adobe Stock and public domain images
> 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.
Thank you for the info. It's still going to run afoul of the copyright office's interpretation of who owns copyright for AI generated materials, but I'm really glad to see them doing something ethical with the model creation.
So your understanding of the law is that any use of the inpainting tool in modern versions of Photoshop makes the image impossible to copyright? Heck, even the magic-wand select is using AI to fine tune its behavior.
Not the entire image, but parts created by inpainting. Which is admittedly a hot mess of red tape in potentia. But it makes sense from the point of view of copyright having been created to protect people's ability to make a living from IP.
The magic wand is different IMO, because it's not creating any imagery.
If any if the training set is public domain or creative commons, I'd argue that this make the image at best an integration work, not derivative nor fair use, and as bits from the images are indistinguishable from each other, you shouldn't be able to license it.
Like code generated from models trained on AGPL/GPLv3 licensed code should also be AGPL/GPLv3.
Nature photographers can spend weeks or months scouting out a proper locale, picking the vantage point, waiting for just the right moment (or collecting and curating hundreds of moments), and then doing (possibly) final touch-ups in a photo editor.
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.
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.
> “The refusal of the U.S. Copyright Office to recognize human authorship in AI-assisted creations highlights a critical issue in modern intellectual property law. As AI continues to evolve, it is imperative that our legal frameworks adapt to protect the rights of those who harness these technologies for creative expression,” Allen’s lawyer, Pester, recently said.
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.
I don't agree that AI generated work should grant a copyright to the prompter, but it's not an inconsistent position. The belief that training a model is fair use does not preclude the belief that copying the work directly is not.
Real AI artists are spending hundreds of hours and their human labor is absolutely copyrightable.
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:
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.
Not really - they've ruled that the output of those flows may not be protected in some cases, but the actual node based comfy-ui flow is almost certainly something you can protect.
I don't think I agree. I know a woman who live by copying paintings (huge ones usually, only saw her copy 2-6m wide painting) in different museums, then sell the copy to buyers who want them in his manor. She doesn't copy 1 for 1 (it's impossible to get all the brush strokes right, even though she spend days studying them), and in fact specialize in 3 specific styles to avoid visible divergence, but her work isn't copyrightable. She spend arguably way, waaaaaay more time on her paintings than 'ai artists' (I met her in a art museum in Lille where she was halfway through the painting after 2 months, she became fast friend with my dad though, spent hours speaking about pigment and brush techniques), that doesn't make her work any more copyrighted.
I don't necessarily disagree, but I don't necessarily agree either.
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"...
It's about who created the work (book, painting, code, etc). If it's a human, then it's copyrightable. If it's an AI, then it's not. That's the stance the copyright office has been taking, going so far as to grant partial copyrights if a human substantially changes portions of an AI generated work.
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.
The U.S. Copyright Office has made a bogus claim -- 'work derived from AI platforms “contained no human authorship”' -- my belief is that this incorrect and probably illegal edict is simply a poor excuse for the office to neglect its mandate. They do not have the resources and technological footing to register the flood of AI art they would likely receive for copyright protection. Regardless of chutzpah I support Allen's move against it. Style has not been copyrightable, living human artists learn and are influenced from their exposure to the broad history of art, it is not somehow different for AI models to do so as well. Infringement lies in the output of AI models, not on its inputs. Artists using generative art making techniques have been provided copyright protections for generations. While inconvenient for the office, AI mediated works also warrant protection.
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