If Disney wins this, it won't destroy AI image generation as a whole, only models like Midjourney and Stable Diffusion trained on the entire web, limiting access to companies large enough to own...
If Disney wins this, it won't destroy AI image generation as a whole, only models like Midjourney and Stable Diffusion trained on the entire web, limiting access to companies large enough to own or license their own training data. So all the downsides of AI media, but with bonus mandatory corporate gatekeeping and rent-seeking.
AI isn't going away, so the best outcome is distributing its power as widely as possible instead of concentrating power even further. IMHO, that means ruling that any AI media is, by definition, public domain. From the Commons it comes, and to the Commons it must return.
I think there have already been two (?) cases that have ruled along those lines, using the monkey precedent. There isn't sufficient effort under the "modicum of creativity" doctrine (which is...
I think there have already been two (?) cases that have ruled along those lines, using the monkey precedent. There isn't sufficient effort under the "modicum of creativity" doctrine (which is barely any) to consider the operator of the generator to be a copyright holder, and only a human can hold copyright (reasonably, if a machine is entitled to human rights, it cannot have an owner), so the result is public domain
The odd part is that the monkey selfie case ruled the photo to be the work of the monkey because it pressed the shutter button. No mention of the CMOS sensor and associated processing chips that...
The odd part is that the monkey selfie case ruled the photo to be the work of the monkey because it pressed the shutter button. No mention of the CMOS sensor and associated processing chips that created the actual image.
One would think that writing a prompt, causing a silicon chip to create an image, takes at least marginally more effort and creativity than pressing a button…
(Not a value judgement either way, btw - I do have a lot of strong opinions there, but I’ll save them for another time. This is more a question of inconsistency regardless of value judgement)
Assuming I understand the case correctly this is slightly incorrect, but in a very significant way. The court ruling did not say that the monkey has the copyright, but instead that no one has the...
Assuming I understand the case correctly this is slightly incorrect, but in a very significant way. The court ruling did not say that the monkey has the copyright, but instead that no one has the copyright. This is relevant because it isn't a claim that the monkey exercised significant creativity, but is instead just a claim that no human exercised significant enough creativity.
In other words: "pressing a button" isn't the bar because it wasn't trying to rule on how creative pressing a button is. It was ruling that being the owner of equipment also does not meet the bar.
The same can hold true with AI: it's possible for works to be created where no human put enough creativity into it for anyone to hold the copyright. It's widely believed that prompting is probably not enough, but to my knowledge that hasn't been challenged. There's also probably a sliding scale that exists though it is likely too complex to be exercised in reality (ex. typing "cat" is at one end where the other end may be hours or days refining prompts and masks and inpainting and whatever else).
[Edit] The contradiction I’m seeing from the copyright office really bothers me, so I’d actually be happier if someone could make a convincing argument for why snapshot photography or footage from...
[Edit] The contradiction I’m seeing from the copyright office really bothers me, so I’d actually be happier if someone could make a convincing argument for why snapshot photography or footage from unattended, automated cameras has sufficient creative input, but a written prompt does not.
The votes make it pretty clear that people disagree with my take, but I genuinely don’t see the gap in the logic. I’d appreciate anyone’s thoughts on how the precedent for photography (monkey case, security cameras, etc.) suggests that more creative effort than writing a prompt is required?
My understanding is that the photographer’s argument is that he did exercise significant creativity, by taking every step prior to the actual button press, and that the last trivial step was the only one undertaken by the monkey. Rejecting that claim means the button press is considered to hold at least equal significance to all of the adjacent preparation and setup work - assuming we’re taking this as precedent at all, since there were a few different suits filed and I believe the most interesting one was settled, meaning it wasn’t ultimately ruled on.
So you’re techically correct, the monkey selfie case tells us what can’t be copyrighted rather than what can. But it means that by definition the button press is the legally significant part of the process - otherwise the photographer would hold copyright as a result of all the creative work and input taken to engineer the situation prior to that final point. The monkey’s only in the running if pressing the shutter button can confer copyright.
This is also reinforced by the fact that I can point my phone out the window with zero thought or preparation and will still hold copyright on the resulting image, but that’s a much less fun line of thinking to work through!
Photography is a simple cut and dry when it comes to the origin of the creator. If the monkey hadn't pressed the button, the work wouldn't exist at all. Therefore, it was the monkey's decision to...
Photography is a simple cut and dry when it comes to the origin of the creator. If the monkey hadn't pressed the button, the work wouldn't exist at all. Therefore, it was the monkey's decision to create the work, not the guy, even if he owned the equipment. My point being, unless you, a human, make a decision to make a thing, there's no reason you would ever hold the copyright.
Diffusion generated images have a similar button push that makes their works exist.
You push a button to start recording out your window.
The only settled case was whether a non human can hold copyright, which was still picked up by a court and ruled that no, in America, only humans can hold copyright.
Honestly I wish it were cut and dry, it’d make it easier to discuss the contradictions in what the copyright office are saying! It does sound like we’re in agreement on the similarity between...
Honestly I wish it were cut and dry, it’d make it easier to discuss the contradictions in what the copyright office are saying! It does sound like we’re in agreement on the similarity between taking a photo and promoting a diffusion model, at least, which is reassuring.
On the photography side there are situations like security cameras with timed recording, trail cameras that record based on motion, etc. that all still confer copyright on the person who set them up - that’s one of the reasons I have a strong opinion on the “button” part of the situation.
Comparing those examples to:
Slater went on to say, "I put my camera on a tripod with a very wide angle lens, settings configured such as predictive autofocus, motorwind, even a flashgun, to give me a chance of a facial close up if they were to approach again for a play ... I had one hand on the tripod when this was going on, but I was being prodded and poked by would be groomers and a few playful juveniles who nibbled at my arms." In a November 2017 interview with the radio show This American Life, Slater said that he was holding the tripod with his fingers when the images were taken.
I can’t see any conclusion to draw here other than the button press itself is more important than all of those prior actions in the eyes of the law, especially since he would have unquestionably held copyright if he’d just disabled the button and used a timer or IR trigger instead.
If I line up a shot with the button under a pendulum does the clock get the copyright? A non-human pressing something doesn't necessarily mean it expressed agency like you're saying. However, I do...
If I line up a shot with the button under a pendulum does the clock get the copyright? A non-human pressing something doesn't necessarily mean it expressed agency like you're saying. However, I do think monkeys could be trained to take photographs and might use their agency to photograph things they like.
The following is my effort to answer your edit in particular. I'm also not trying to argue how copyright should be by instead focusing on my understanding of how copyright is. My understanding of...
The following is my effort to answer your edit in particular. I'm also not trying to argue how copyright should be by instead focusing on my understanding of how copyright is.
My understanding of the ruling suggests it's about a concept I'll call "creative intent". I'll define this as "intentionally creating a work that is comprised of a significant proportion of unique content where the uniqueness is derived from creative decisions." I'm sure it could be defined better, but I'm not good at this kind of stuff so that'll suffice for now. This also assumes some things that I understand to be true but could be wrong about, such as that accidentally turning on a webcam doesn't give you the copyright to the video even if it's your webcam filming you.
The "significant portion of unique content" is intentionally ambiguous because copyright is fuzzy. For example, you famously cannot copyright a chord or even a short chord progression, but eventually there is a long enough string of them that you can copyright it. How long exactly that must be is a more complex topic. Pointing a camera out the window and pressing the button is less effort, but the resulting work is going to be substantially unique in a way that is derived from the decision to do so (that window, that time, that angle, that camera, etc.).
How this relates to the monkey case is that, at least as far as I can tell, he intended to get pictures of monkeys but was not the one exercising intent when the photos were actually taken. In a similar vein with humans only, if the monkey was a very young human child that hit the button would it still be fair to say the owner of the camera should have the copyright? The child would have the claim that it wasn't the owner's intent to have the button pressed at the moment it was, and the owner would have the argument that they intentionally brought the equipment and using it wasn't the child's intent so only the intent to make it possible should matter. Both sides undercut the other's intent and I think that's the direction the ruling was coming from. If the camera had been left on in some way, such as in video mode or set to snap a picture every second, I'd assume the ruling would've been very different because "I deliberately put the camera into an automated mode and gave it to a monkey" has an easier intent to argue.
How this relates to AI is mostly just that it clarifies that setting off the chain of events that lead to the work being created does not in itself show enough creative intent. As a concrete example, if I told an LLM "write a funny story" the tiny bit of effort I put in is not even substantiality represented by the produced work because any uniqueness relative to any other funny story was not caused by me. A fun thought experiment about what it would mean if I could is if I had the computational power and told a computer "generate every syntactically valid combination of up to 1 million words". Should it really be granted that I then just own basically every written work that didn't previously exist just because I "pushed the button" telling a computer to exercise my idea to copyright everything?
From my perspective, the monkey case is one of a person who absolutely did exercise significant creative intent, which I think is a big part of why the apparent contradiction bothers me: Taking it...
From my perspective, the monkey case is one of a person who absolutely did exercise significant creative intent, which I think is a big part of why the apparent contradiction bothers me:
Slater went on to say, "I put my camera on a tripod with a very wide angle lens, settings configured such as predictive autofocus, motorwind, even a flashgun, to give me a chance of a facial close up if they were to approach again for a play ... I had one hand on the tripod when this was going on, but I was being prodded and poked by would be groomers and a few playful juveniles who nibbled at my arms." In a November 2017 interview with the radio show This American Life, Slater said that he was holding the tripod with his fingers when the images were taken.
Taking it a step further and looking outside that case, there's a lot of precedent showing that the actual amount of creative intent required is, bluntly put, almost zero. A direct quote from one of the key test cases on the subject says as much fairly explicitly:
To be sure, the requisite level of creativity is extremely low; even a slight amount will suffice. The vast majority of works make the grade quite easily, as they possess some creative spark, "no matter how crude, humble or obvious" it might be.
My line of thinking is nothing to to with him being the owner of the camera, it's to do with him setting up every creative part of the process, clearly and deliberately, except for that single button press. If it were a human child who hit the button while waving their arms and playing around, after the photographer had set up the camera and tripod, checked the lighting, and framed things ready to capture a photo, I'd be saying the exact same thing.
How this relates to AI is mostly just that it clarifies that setting off the chain of events that lead to the work being created does not in itself show enough creative intent.
This is the part I can't reconcile. Doing the hard, high effort, creative part without pressing the button apparently isn't enough. But if I take my phone out, make a split second decision to point it at something, and then hit the shutter button (which then starts a process where the software, CMOS sensor, and image processing algorithms work to bring a picture into existence) there's absolutely no argument that I have copyright over the image.
I can unequivocally spend less effort on a photo than it would take me to think up a six word prompt, then press a button to trigger a computational process in which hardware and software works together to create that photo, and have copyright over that photo. It just doesn't seem consistent to me that prompting a diffusion model isn't equal to that very low bar, and I can't find a way to look at it that it makes sense.
And for complete clarity here, I'm not saying this in favour of copyrighting AI generated images, I'm saying it as a way to highlight how broken and nonsensical copyright law has already become - it's absolutely not fit for purpose as it is, let alone fit for handling the nuances of generative algorithms.
It is possible, though, to obtain a copyright in a work that is identical to an earlier work, so long as the author did not copy from the earlier work, either consciously or subconsciously. Unlike patent law, copyright law does not require novelty. Judge Learned Hand gave an example of this principle in the 1936 case, Sheldon v. Metro-Goldwyn: “Borrowed the work must indeed not be, . . . but if by some magic a man who had never known it were to compose anew Keats's Ode on a Grecian Urn, he would be an ‘author,’ and . . . others might not copy that poem, though they might of course copy Keats's.”
I think copyright law could apply to generative AI similarly. Copyright ownership depends not just on creative effort and intent but also on the subject. If you happened to take a snapshot of...
I can point my phone out the window with zero thought or preparation and will still hold copyright on the resulting image
I think copyright law could apply to generative AI similarly. Copyright ownership depends not just on creative effort and intent but also on the subject. If you happened to take a snapshot of someone else's copyrighted work (for ex. a Disney movie billboard, or the Eiffel Tower) then you wouldn't own the copyrights to your snapshot.
Same if you used generative AI to produce output that (intentionally or not) highly resembled someone else's copyrighted work. It could also be argued that if the generative AI tool itself was created by training on a lot of (not properly licensed) copyrighted material then all its output's copyright could be questionable. This wouldn't be the case with past mediums, for ex. collages. But the law should adapt to new technology, and this technology does enable infringement (and obscurement of original authorship) on a massive and unprecedented scale so I think the argument is valid.
Copyright law is supposed to protect creators and reward them for their work; generative AI actually does the opposite. It makes sense to me to use copyright to fight this.
And just because big corporations like Disney have been harmful to artists, doesn't give generative AI a free pass to also be harmful to artists.
That said I can imagine conditions under which the use of generative AI would be ethically ok and its output copyrightable:
trained using public domain and/or properly licensed material
if public domain used for training then the generative AI tool must be kept open source-- a big corporation should not be allowed to take from the public domain without giving back, but it's ok for individual creators to use the tool to create their own copyrightable work
I like your broad line of thinking here, and it's fairly similar to where I end up on the bigger picture questions. There would definitely be fine details to hash out, but this kind of sensible,...
I like your broad line of thinking here, and it's fairly similar to where I end up on the bigger picture questions. There would definitely be fine details to hash out, but this kind of sensible, creator-focused conversation about reforming a system that's no longer keeping up with new developments is exactly what we need.
One nitpick, though, because I think the fine nuances are a big part of what makes this whole situation interesting: you do own the copyright to a photo you take of the Eiffel tower lighting, or a Disney billboard, or similar, at least if there's any element of framing or composition at all there. There may be limitations on what you can do with your photo because it also contains a representation of someone else's copyrighted work, but the photo itself is still yours. The implications of that could well be significant when it comes to understanding where generative models fit in under current copyright law, even if it doesn't really change the broader question of what reforms would be sensible.
That makes a lot of sense to me and wouldn't prevent companies like Disney from using AI for private ventures. Making these things public domain would effectively socialize the benefits of...
That makes a lot of sense to me and wouldn't prevent companies like Disney from using AI for private ventures. Making these things public domain would effectively socialize the benefits of generated images, which is reasonable when their makeup was sourced by all of society. If data has a dollar value, you could say that these companies owe the pubic a significant stake in their value.
Yeahhhhh. It's an article so i'm suspicious, but I hope for their sake their defense is better than that. Granted it's 100% the truth, but it's also not a valid defense. The two points of this...
“There isn’t really a way to get a hundred million images and know where they’re coming from. It would be cool if images had metadata embedded in them about the copyright owner or something. But that’s not a thing; there’s not a registry. There’s no way to find a picture on the Internet, and then automatically trace it to an owner and then have any way of doing anything to authenticate it.”
Yeahhhhh. It's an article so i'm suspicious, but I hope for their sake their defense is better than that. Granted it's 100% the truth, but it's also not a valid defense.
The two points of this are:
Everyone else is stealing too! Which doesn't matter when you become the sole source of stolen content. Lots of laws aren't enforced in every possible case, and it's recognized that's just a reality, but it's also not a good argument for when they do bust you. Clearly most companies don't go after people who pirate media, but they DO go after the larger distributors, which AI basically became.
We can't tell we stole this! Which is even worse than the previous one. They very likely scraped properly documented and copyrighted material. Some media company with a picture of thanos on their site that they were APPROVED BY DISNEY to use is certainly in their training data.
Naturally there's no way Disney wants AI to go away, they just want to make sure the normal people can't use it, and it looks like midjourney is set to give them the case they need to do so. "oh this model was built with all of our data, so only people who pay us for it can use it" is sadly going to be a very valid argument for all this.
I think a good faith interpretation would be to assume you don’t have rights to the images by default. At least in the US images are automatically under copyright. Very few would be permissively...
I think a good faith interpretation would be to assume you don’t have rights to the images by default. At least in the US images are automatically under copyright. Very few would be permissively licensed.
Overall my reaction is “… let them fight”. But I more so want Disney to lose just because it feels like the bigger blow to the owner class.
I think many of the people who are the most opposed to generative AI practices also really don't want to open the copyright can of worms they're running headlong toward. Namely, independent...
I think many of the people who are the most opposed to generative AI practices also really don't want to open the copyright can of worms they're running headlong toward. Namely, independent artists who draw commissioned art for fans of various media franchises. What they do is concretely copyright infringement, under the modern draconian form.
If you take money to make a depiction of Hatsune Miku or John Wick or whatever, it's very definitely copyright infringement, on a commercial level. I think most people believe this is something artists should be able to do, and it's certainly prevalent in fan communities, but it's decidedly not legal.
Music already got shot in the knee by the courts, during the "Sample Wars," and put a chilling effect on many genres of music. As the technology to detect very short samples, even after they've been repitched, distorted and transformed, becomes more commonplace, we're likely to see another apocalypse of 25+ year old dance hits being soaked for settlements due to them using a 0.5 second sample for an orchestra stab. (Or maybe instrument makers will decide they own the timbre of your instrument, and you didn't pay a license fee before releasing a song with it...)
Many other normalized things people do every day, like putting text over a still frame from a TV show and posting it on social media, are also copyright infringement.
Any copyright-based legal cudgel used against generative AI will come back around to harm independent artists of any medium, and passive participators in pop culture, tenfold.
I dislike the zero-effort spam these things generate, and the impolite crawlers the companies are using, and many other aspects of LLMs and image generators, but copyright is not the answer. Copyright is an evil to be reduced, not empowered. It's a plague on the freedom to make music and visual art, and of little use to the vast majority of artists who will never be "commercially successful," all to benefit a relatively small class of landlords of imaginary property.
It's true, they're already technically operating outside the law. But they operate at such a small scale that copyright holders generally ignore them. I don't think a win against AI companies...
What they do is concretely copyright infringement, under the modern draconian form.
It's true, they're already technically operating outside the law. But they operate at such a small scale that copyright holders generally ignore them. I don't think a win against AI companies would lead to Disney going after small time artists posting fan art on social media, it's not really in their best interests even if it is within their legal rights.
The thing is it's drawing attention to them. What feeds things like stable diffusion models? Not merely official Disney art, which is a relatively small data set, but massive public fan art dumps...
The thing is it's drawing attention to them. What feeds things like stable diffusion models? Not merely official Disney art, which is a relatively small data set, but massive public fan art dumps like Danbooru (which is why the earlier versions skewed NSFW by default).
If you do legal discovery on training materials, it will become obvious that many of the infringing examples are not by Disney but by other artists already infringing upon Disney. And modern tools, and lawsuits against web platforms those artists may inhabit, can definitely turn many "too small" actors into a single, tractable problem.
Recall the Viacom lawsuit against YouTube, which lead to ContentID. Then imagine such a system, operating on stills as well as video, being inflicted upon every major social media platform and any other site that allowed user submitted content. All you have to do is sue the platforms to hurt the artists.
Additionally, it's further precedent strengthening the concept of derivative works, when we should be working to eliminate that concept. Copyright should ideally protect one work, and anything that's substantively close to identical, not create an intellectual fiefdom.
I'll just add that Danbooru is used extremely broadly because it has labels. Labelling images is a huge amount of labor which is not compensated by these thieves. Before OpenAI stole everything...
I'll just add that Danbooru is used extremely broadly because it has labels. Labelling images is a huge amount of labor which is not compensated by these thieves.
Before OpenAI stole everything that wasn't nailed down, there were established labeled datasets to work with. Sam Altman is the villain of all of this, Google would have gotten us where we are without the absolute disregard for intellectual property and the environment.
Funny you mention Hatsune Miku specifically, since a friend recently informed me that her character design is actually licensed for non-commercial use, which means you're legally allowed to use...
Funny you mention Hatsune Miku specifically, since a friend recently informed me that her character design is actually licensed for non-commercial use, which means you're legally allowed to use her design for fan art and such if you don't make money off it. They do explicitly state that you need permission to make money off of the design or use it for a business purpose, which many people definitely do anyways. https://piapro.net/intl/en_for_creators.html
Copyright definitely lasts way too long, and we can see in fantastic unofficial fan projects what we could have with reduced copyright, as opposed to the very lackluster official remakes or sequels we get from copyright holders. At the very least, it'd be nice if more iconic characters went the Miku route allowing non-commercial fan art officially.
Yep, I'm actually very knowledgeable of all things Miku, and the particulars of music releases lol. (The reason is an exercise left to the reader.) The original Miku art is Creative Commons CC-NC...
Yep, I'm actually very knowledgeable of all things Miku, and the particulars of music releases lol. (The reason is an exercise left to the reader.)
The original Miku art is Creative Commons CC-NC in the west, and in Japan they use their own similarly permissive (but more specific) license.
It's unfortunately not super cut and dry, as "commercial activity" could be considered to include: being on a major music streaming service, or charging money to make artwork for a video release, so it's still kind of at the whim of the copyright holder. Word of Crypton is that they consider any YouTube/NicoNico/social media stuff as "doujin activity" and not commercial. (As is selling craft merch at something like Comiket.) So the actual releasing of that to the public is mostly in the clear (because people did it, technically infringing, for years and it built their whole empire up...but the world now makes it harder for something like that to happen again).
When it comes to releasing music on a streaming service, things are more complicated. Because Crypton does consider that commercial, and distributors will ask for proof of rights. In the west, the norm is to not put "Hatsune Miku" in the metadata or artwork and to not use any related imagery. In Japan, they have a contact form you can use to request permission and receive a license.
Realistically, they make their money on licensing Miku for commercial goods, far more than they ever would just selling the Vocaloid voice bank, so it at least makes sense they're sort of careful on that front.
It's not a perfect situation, but at least Crypton is a brilliant case of attempting to do things right.
Edit: Thanks for drawing attention to that. I kind of wanted to, but didn't want my inevitable digression to consume my first comment entirely.
It's vastly more complicated because the reality is closer to "modern copyright law just doesn't properly map to reality in the slightest, and no one has the solution or inclination to figure one...
I think a good faith interpretation would be to assume you don’t have rights to the images by default. At least in the US images are automatically under copyright. Very few would be permissively licensed.
It's vastly more complicated because the reality is closer to "modern copyright law just doesn't properly map to reality in the slightest, and no one has the solution or inclination to figure one out." This court case somewhat forces that issue finally.
Overall my reaction is “… let them fight”. But I more so want Disney to lose just because it feels like the bigger blow to the owner class.
Realistically, this is not a very useful view. Disney "losing" isn't harming them or the owner class in the slightest.
I think you want Disney to win this. Disney is terrible, for sure. But Trump is worse. And the Trump admin loves AI (I'm sure there are various backdoor bribes and other grifts going on). The...
I think you want Disney to win this.
Disney is terrible, for sure. But Trump is worse. And the Trump admin loves AI (I'm sure there are various backdoor bribes and other grifts going on).
The article mentions how Trump fired the head of the copyright office and how the "Big Bullshit Bill" has protections for AI.
In case it isn't obvious, I know how bad Disney has been for artists in the past, but at least they attempt to follow the law, even though I also understand that they use a lot of influence to shape it in their favor. With the backing of the lawless Trump admin and republican party, AI is going to get even more completely out of control. For both artists and the population in general.
You see, what we do know is that after the Copyright Office released a pre-publication version of its 108-page copyright and AI report, which strived to strike a middle ground “by supporting both of these world-class industries that contribute so much to our economic and cultural advancement.” However, it added that while some generative AI probably constitutes a “transformative” use, the mass scraping of all data did not qualify as fair use.
The result? The Trump administration, while not commenting on the report, fired Shira Perlmutter, the head of the Copyright Office, the next day. She’s been replaced by an attorney with no IP experience.
Oh, also, hidden away in Trump’s “One Big Beautiful Bill” is a statement that imposes a 10-year ban on the enforcement of any state or local laws or regulations that “limit, restrict, or otherwise regulate” AI models, AI systems, or automated decision systems. If that becomes law, whatever is in Trump’s AI Action Plan is what we’ll have to live with for the next few years.
Assuming that you don’t have the rights does imply that training on an image is definitely copying under the definitions of current copyright law, though. I don’t think that’s anywhere near clear...
Assuming that you don’t have the rights does imply that training on an image is definitely copying under the definitions of current copyright law, though. I don’t think that’s anywhere near clear enough to be a given - I’m with @Eji1700 on this part, we’re off the map in territory that copyright law just doesn’t account for yet.
Overall my reaction is “… let them fight”. But I more so want Disney to lose just because it feels like the bigger blow to the owner class.
They're right in that it's kind of unfeasible verify the provenance of all data in such large training sets. Unfortunately for them, I (and I think most consumers) don't care if enforcing...
They're right in that it's kind of unfeasible verify the provenance of all data in such large training sets. Unfortunately for them, I (and I think most consumers) don't care if enforcing copyright law makes the business models of these shitty AI companies completely non-viable. I don't see how making an exception for them would benefit society as a whole.
I never thought I'd say this before, but I hope Disney wins this copyright suit.
Yeah from the point of view of the law you're basically saying: "well we've gotten so good at hoovering up random crap we can't tell if any of it is stolen", which, yeah ok. Doesn't matter? It...
Yeah from the point of view of the law you're basically saying:
"well we've gotten so good at hoovering up random crap we can't tell if any of it is stolen", which, yeah ok. Doesn't matter? It means you've admitted you know you're stealing and it can be proved you're stealing. The fact you came up with a new way to collect loose crap doesn't change that fact, nor does it make it legal to do so.
I think there's a difference between having an artistic capability (it could generate an Iron Man image) and actually doing it. If the buyer explicitly names the character in the prompt, maybe...
I think there's a difference between having an artistic capability (it could generate an Iron Man image) and actually doing it. If the buyer explicitly names the character in the prompt, maybe it's their fault for asking for such a thing?
If it's generating such images for a generic prompt, that's quite another thing.
A simple Google image search looking for an existing thing will pull AI generated images without anyone inputting anything For example, I saw a logo/initials as a sticker on someone's car that was...
A simple Google image search looking for an existing thing will pull AI generated images without anyone inputting anything
For example, I saw a logo/initials as a sticker on someone's car that was KHK but the H looked very much "K" like due to the design (a slanted cross bar for example). So I tried to Google (like I've done in the past) for "KHK logos" "KHK initials looks like KKK" and a number of other iterations. The image search was full of helpful and as far as I can tell newly generated logos for me for the KKK.
I played with my settings but never found the exact thing I'm looking for. If I removed AI images I only got pics related to a Qatari (I think) company with the same name. So while maybe they're keyword blocking this from Disney itself I haven't tested, it's actually harder not to get forcefed AI images than not.
Sometimes the big companies are just plain at fault.
I think the issue is more with training data than output. Unauthorized use is still unauthorized, and copyright sucks with how restrictive it is in many countries.
I think the issue is more with training data than output. Unauthorized use is still unauthorized, and copyright sucks with how restrictive it is in many countries.
If Disney wins this, it won't destroy AI image generation as a whole, only models like Midjourney and Stable Diffusion trained on the entire web, limiting access to companies large enough to own or license their own training data. So all the downsides of AI media, but with bonus mandatory corporate gatekeeping and rent-seeking.
AI isn't going away, so the best outcome is distributing its power as widely as possible instead of concentrating power even further. IMHO, that means ruling that any AI media is, by definition, public domain. From the Commons it comes, and to the Commons it must return.
I think there have already been two (?) cases that have ruled along those lines, using the monkey precedent. There isn't sufficient effort under the "modicum of creativity" doctrine (which is barely any) to consider the operator of the generator to be a copyright holder, and only a human can hold copyright (reasonably, if a machine is entitled to human rights, it cannot have an owner), so the result is public domain
The odd part is that the monkey selfie case ruled the photo to be the work of the monkey because it pressed the shutter button. No mention of the CMOS sensor and associated processing chips that created the actual image.
One would think that writing a prompt, causing a silicon chip to create an image, takes at least marginally more effort and creativity than pressing a button…
(Not a value judgement either way, btw - I do have a lot of strong opinions there, but I’ll save them for another time. This is more a question of inconsistency regardless of value judgement)
Assuming I understand the case correctly this is slightly incorrect, but in a very significant way. The court ruling did not say that the monkey has the copyright, but instead that no one has the copyright. This is relevant because it isn't a claim that the monkey exercised significant creativity, but is instead just a claim that no human exercised significant enough creativity.
In other words: "pressing a button" isn't the bar because it wasn't trying to rule on how creative pressing a button is. It was ruling that being the owner of equipment also does not meet the bar.
The same can hold true with AI: it's possible for works to be created where no human put enough creativity into it for anyone to hold the copyright. It's widely believed that prompting is probably not enough, but to my knowledge that hasn't been challenged. There's also probably a sliding scale that exists though it is likely too complex to be exercised in reality (ex. typing "cat" is at one end where the other end may be hours or days refining prompts and masks and inpainting and whatever else).
[Edit] The contradiction I’m seeing from the copyright office really bothers me, so I’d actually be happier if someone could make a convincing argument for why snapshot photography or footage from unattended, automated cameras has sufficient creative input, but a written prompt does not.
The votes make it pretty clear that people disagree with my take, but I genuinely don’t see the gap in the logic. I’d appreciate anyone’s thoughts on how the precedent for photography (monkey case, security cameras, etc.) suggests that more creative effort than writing a prompt is required?
My understanding is that the photographer’s argument is that he did exercise significant creativity, by taking every step prior to the actual button press, and that the last trivial step was the only one undertaken by the monkey. Rejecting that claim means the button press is considered to hold at least equal significance to all of the adjacent preparation and setup work - assuming we’re taking this as precedent at all, since there were a few different suits filed and I believe the most interesting one was settled, meaning it wasn’t ultimately ruled on.
So you’re techically correct, the monkey selfie case tells us what can’t be copyrighted rather than what can. But it means that by definition the button press is the legally significant part of the process - otherwise the photographer would hold copyright as a result of all the creative work and input taken to engineer the situation prior to that final point. The monkey’s only in the running if pressing the shutter button can confer copyright.
This is also reinforced by the fact that I can point my phone out the window with zero thought or preparation and will still hold copyright on the resulting image, but that’s a much less fun line of thinking to work through!
Photography is a simple cut and dry when it comes to the origin of the creator. If the monkey hadn't pressed the button, the work wouldn't exist at all. Therefore, it was the monkey's decision to create the work, not the guy, even if he owned the equipment. My point being, unless you, a human, make a decision to make a thing, there's no reason you would ever hold the copyright.
Diffusion generated images have a similar button push that makes their works exist.
You push a button to start recording out your window.
The only settled case was whether a non human can hold copyright, which was still picked up by a court and ruled that no, in America, only humans can hold copyright.
Honestly I wish it were cut and dry, it’d make it easier to discuss the contradictions in what the copyright office are saying! It does sound like we’re in agreement on the similarity between taking a photo and promoting a diffusion model, at least, which is reassuring.
On the photography side there are situations like security cameras with timed recording, trail cameras that record based on motion, etc. that all still confer copyright on the person who set them up - that’s one of the reasons I have a strong opinion on the “button” part of the situation.
Comparing those examples to:
I can’t see any conclusion to draw here other than the button press itself is more important than all of those prior actions in the eyes of the law, especially since he would have unquestionably held copyright if he’d just disabled the button and used a timer or IR trigger instead.
If I line up a shot with the button under a pendulum does the clock get the copyright? A non-human pressing something doesn't necessarily mean it expressed agency like you're saying. However, I do think monkeys could be trained to take photographs and might use their agency to photograph things they like.
The following is my effort to answer your edit in particular. I'm also not trying to argue how copyright should be by instead focusing on my understanding of how copyright is.
My understanding of the ruling suggests it's about a concept I'll call "creative intent". I'll define this as "intentionally creating a work that is comprised of a significant proportion of unique content where the uniqueness is derived from creative decisions." I'm sure it could be defined better, but I'm not good at this kind of stuff so that'll suffice for now. This also assumes some things that I understand to be true but could be wrong about, such as that accidentally turning on a webcam doesn't give you the copyright to the video even if it's your webcam filming you.
The "significant portion of unique content" is intentionally ambiguous because copyright is fuzzy. For example, you famously cannot copyright a chord or even a short chord progression, but eventually there is a long enough string of them that you can copyright it. How long exactly that must be is a more complex topic. Pointing a camera out the window and pressing the button is less effort, but the resulting work is going to be substantially unique in a way that is derived from the decision to do so (that window, that time, that angle, that camera, etc.).
How this relates to the monkey case is that, at least as far as I can tell, he intended to get pictures of monkeys but was not the one exercising intent when the photos were actually taken. In a similar vein with humans only, if the monkey was a very young human child that hit the button would it still be fair to say the owner of the camera should have the copyright? The child would have the claim that it wasn't the owner's intent to have the button pressed at the moment it was, and the owner would have the argument that they intentionally brought the equipment and using it wasn't the child's intent so only the intent to make it possible should matter. Both sides undercut the other's intent and I think that's the direction the ruling was coming from. If the camera had been left on in some way, such as in video mode or set to snap a picture every second, I'd assume the ruling would've been very different because "I deliberately put the camera into an automated mode and gave it to a monkey" has an easier intent to argue.
How this relates to AI is mostly just that it clarifies that setting off the chain of events that lead to the work being created does not in itself show enough creative intent. As a concrete example, if I told an LLM "write a funny story" the tiny bit of effort I put in is not even substantiality represented by the produced work because any uniqueness relative to any other funny story was not caused by me. A fun thought experiment about what it would mean if I could is if I had the computational power and told a computer "generate every syntactically valid combination of up to 1 million words". Should it really be granted that I then just own basically every written work that didn't previously exist just because I "pushed the button" telling a computer to exercise my idea to copyright everything?
From my perspective, the monkey case is one of a person who absolutely did exercise significant creative intent, which I think is a big part of why the apparent contradiction bothers me:
Taking it a step further and looking outside that case, there's a lot of precedent showing that the actual amount of creative intent required is, bluntly put, almost zero. A direct quote from one of the key test cases on the subject says as much fairly explicitly:
My line of thinking is nothing to to with him being the owner of the camera, it's to do with him setting up every creative part of the process, clearly and deliberately, except for that single button press. If it were a human child who hit the button while waving their arms and playing around, after the photographer had set up the camera and tripod, checked the lighting, and framed things ready to capture a photo, I'd be saying the exact same thing.
This is the part I can't reconcile. Doing the hard, high effort, creative part without pressing the button apparently isn't enough. But if I take my phone out, make a split second decision to point it at something, and then hit the shutter button (which then starts a process where the software, CMOS sensor, and image processing algorithms work to bring a picture into existence) there's absolutely no argument that I have copyright over the image.
I can unequivocally spend less effort on a photo than it would take me to think up a six word prompt, then press a button to trigger a computational process in which hardware and software works together to create that photo, and have copyright over that photo. It just doesn't seem consistent to me that prompting a diffusion model isn't equal to that very low bar, and I can't find a way to look at it that it makes sense.
And for complete clarity here, I'm not saying this in favour of copyrighting AI generated images, I'm saying it as a way to highlight how broken and nonsensical copyright law has already become - it's absolutely not fit for purpose as it is, let alone fit for handling the nuances of generative algorithms.
As for the though experiment, it is a fun one - and not just a thought experiment, in fact! But copyright law already has a neat workaround for that:
I think copyright law could apply to generative AI similarly. Copyright ownership depends not just on creative effort and intent but also on the subject. If you happened to take a snapshot of someone else's copyrighted work (for ex. a Disney movie billboard, or the Eiffel Tower) then you wouldn't own the copyrights to your snapshot.
Same if you used generative AI to produce output that (intentionally or not) highly resembled someone else's copyrighted work. It could also be argued that if the generative AI tool itself was created by training on a lot of (not properly licensed) copyrighted material then all its output's copyright could be questionable. This wouldn't be the case with past mediums, for ex. collages. But the law should adapt to new technology, and this technology does enable infringement (and obscurement of original authorship) on a massive and unprecedented scale so I think the argument is valid.
Copyright law is supposed to protect creators and reward them for their work; generative AI actually does the opposite. It makes sense to me to use copyright to fight this.
And just because big corporations like Disney have been harmful to artists, doesn't give generative AI a free pass to also be harmful to artists.
That said I can imagine conditions under which the use of generative AI would be ethically ok and its output copyrightable:
I like your broad line of thinking here, and it's fairly similar to where I end up on the bigger picture questions. There would definitely be fine details to hash out, but this kind of sensible, creator-focused conversation about reforming a system that's no longer keeping up with new developments is exactly what we need.
One nitpick, though, because I think the fine nuances are a big part of what makes this whole situation interesting: you do own the copyright to a photo you take of the Eiffel tower lighting, or a Disney billboard, or similar, at least if there's any element of framing or composition at all there. There may be limitations on what you can do with your photo because it also contains a representation of someone else's copyrighted work, but the photo itself is still yours. The implications of that could well be significant when it comes to understanding where generative models fit in under current copyright law, even if it doesn't really change the broader question of what reforms would be sensible.
That makes a lot of sense to me and wouldn't prevent companies like Disney from using AI for private ventures. Making these things public domain would effectively socialize the benefits of generated images, which is reasonable when their makeup was sourced by all of society. If data has a dollar value, you could say that these companies owe the pubic a significant stake in their value.
Yeahhhhh. It's an article so i'm suspicious, but I hope for their sake their defense is better than that. Granted it's 100% the truth, but it's also not a valid defense.
The two points of this are:
Everyone else is stealing too! Which doesn't matter when you become the sole source of stolen content. Lots of laws aren't enforced in every possible case, and it's recognized that's just a reality, but it's also not a good argument for when they do bust you. Clearly most companies don't go after people who pirate media, but they DO go after the larger distributors, which AI basically became.
We can't tell we stole this! Which is even worse than the previous one. They very likely scraped properly documented and copyrighted material. Some media company with a picture of thanos on their site that they were APPROVED BY DISNEY to use is certainly in their training data.
Naturally there's no way Disney wants AI to go away, they just want to make sure the normal people can't use it, and it looks like midjourney is set to give them the case they need to do so. "oh this model was built with all of our data, so only people who pay us for it can use it" is sadly going to be a very valid argument for all this.
I think a good faith interpretation would be to assume you don’t have rights to the images by default. At least in the US images are automatically under copyright. Very few would be permissively licensed.
Overall my reaction is “… let them fight”. But I more so want Disney to lose just because it feels like the bigger blow to the owner class.
I think many of the people who are the most opposed to generative AI practices also really don't want to open the copyright can of worms they're running headlong toward. Namely, independent artists who draw commissioned art for fans of various media franchises. What they do is concretely copyright infringement, under the modern draconian form.
If you take money to make a depiction of Hatsune Miku or John Wick or whatever, it's very definitely copyright infringement, on a commercial level. I think most people believe this is something artists should be able to do, and it's certainly prevalent in fan communities, but it's decidedly not legal.
Music already got shot in the knee by the courts, during the "Sample Wars," and put a chilling effect on many genres of music. As the technology to detect very short samples, even after they've been repitched, distorted and transformed, becomes more commonplace, we're likely to see another apocalypse of 25+ year old dance hits being soaked for settlements due to them using a 0.5 second sample for an orchestra stab. (Or maybe instrument makers will decide they own the timbre of your instrument, and you didn't pay a license fee before releasing a song with it...)
Many other normalized things people do every day, like putting text over a still frame from a TV show and posting it on social media, are also copyright infringement.
Any copyright-based legal cudgel used against generative AI will come back around to harm independent artists of any medium, and passive participators in pop culture, tenfold.
I dislike the zero-effort spam these things generate, and the impolite crawlers the companies are using, and many other aspects of LLMs and image generators, but copyright is not the answer. Copyright is an evil to be reduced, not empowered. It's a plague on the freedom to make music and visual art, and of little use to the vast majority of artists who will never be "commercially successful," all to benefit a relatively small class of landlords of imaginary property.
It's true, they're already technically operating outside the law. But they operate at such a small scale that copyright holders generally ignore them. I don't think a win against AI companies would lead to Disney going after small time artists posting fan art on social media, it's not really in their best interests even if it is within their legal rights.
The thing is it's drawing attention to them. What feeds things like stable diffusion models? Not merely official Disney art, which is a relatively small data set, but massive public fan art dumps like Danbooru (which is why the earlier versions skewed NSFW by default).
If you do legal discovery on training materials, it will become obvious that many of the infringing examples are not by Disney but by other artists already infringing upon Disney. And modern tools, and lawsuits against web platforms those artists may inhabit, can definitely turn many "too small" actors into a single, tractable problem.
Recall the Viacom lawsuit against YouTube, which lead to ContentID. Then imagine such a system, operating on stills as well as video, being inflicted upon every major social media platform and any other site that allowed user submitted content. All you have to do is sue the platforms to hurt the artists.
Additionally, it's further precedent strengthening the concept of derivative works, when we should be working to eliminate that concept. Copyright should ideally protect one work, and anything that's substantively close to identical, not create an intellectual fiefdom.
I'll just add that Danbooru is used extremely broadly because it has labels. Labelling images is a huge amount of labor which is not compensated by these thieves.
Before OpenAI stole everything that wasn't nailed down, there were established labeled datasets to work with. Sam Altman is the villain of all of this, Google would have gotten us where we are without the absolute disregard for intellectual property and the environment.
Funny you mention Hatsune Miku specifically, since a friend recently informed me that her character design is actually licensed for non-commercial use, which means you're legally allowed to use her design for fan art and such if you don't make money off it. They do explicitly state that you need permission to make money off of the design or use it for a business purpose, which many people definitely do anyways. https://piapro.net/intl/en_for_creators.html
Copyright definitely lasts way too long, and we can see in fantastic unofficial fan projects what we could have with reduced copyright, as opposed to the very lackluster official remakes or sequels we get from copyright holders. At the very least, it'd be nice if more iconic characters went the Miku route allowing non-commercial fan art officially.
Yep, I'm actually very knowledgeable of all things Miku, and the particulars of music releases lol. (The reason is an exercise left to the reader.)
The original Miku art is Creative Commons CC-NC in the west, and in Japan they use their own similarly permissive (but more specific) license.
It's unfortunately not super cut and dry, as "commercial activity" could be considered to include: being on a major music streaming service, or charging money to make artwork for a video release, so it's still kind of at the whim of the copyright holder. Word of Crypton is that they consider any YouTube/NicoNico/social media stuff as "doujin activity" and not commercial. (As is selling craft merch at something like Comiket.) So the actual releasing of that to the public is mostly in the clear (because people did it, technically infringing, for years and it built their whole empire up...but the world now makes it harder for something like that to happen again).
When it comes to releasing music on a streaming service, things are more complicated. Because Crypton does consider that commercial, and distributors will ask for proof of rights. In the west, the norm is to not put "Hatsune Miku" in the metadata or artwork and to not use any related imagery. In Japan, they have a contact form you can use to request permission and receive a license.
Realistically, they make their money on licensing Miku for commercial goods, far more than they ever would just selling the Vocaloid voice bank, so it at least makes sense they're sort of careful on that front.
It's not a perfect situation, but at least Crypton is a brilliant case of attempting to do things right.
Edit: Thanks for drawing attention to that. I kind of wanted to, but didn't want my inevitable digression to consume my first comment entirely.
It's vastly more complicated because the reality is closer to "modern copyright law just doesn't properly map to reality in the slightest, and no one has the solution or inclination to figure one out." This court case somewhat forces that issue finally.
Realistically, this is not a very useful view. Disney "losing" isn't harming them or the owner class in the slightest.
I think you want Disney to win this.
Disney is terrible, for sure. But Trump is worse. And the Trump admin loves AI (I'm sure there are various backdoor bribes and other grifts going on).
The article mentions how Trump fired the head of the copyright office and how the "Big Bullshit Bill" has protections for AI.
In case it isn't obvious, I know how bad Disney has been for artists in the past, but at least they attempt to follow the law, even though I also understand that they use a lot of influence to shape it in their favor. With the backing of the lawless Trump admin and republican party, AI is going to get even more completely out of control. For both artists and the population in general.
If so I think we can rest easy, as I would bet money on Disney getting the upper hand here. Maybe not a complete win, though.
Assuming that you don’t have the rights does imply that training on an image is definitely copying under the definitions of current copyright law, though. I don’t think that’s anywhere near clear enough to be a given - I’m with @Eji1700 on this part, we’re off the map in territory that copyright law just doesn’t account for yet.
Strong agree
They're right in that it's kind of unfeasible verify the provenance of all data in such large training sets. Unfortunately for them, I (and I think most consumers) don't care if enforcing copyright law makes the business models of these shitty AI companies completely non-viable. I don't see how making an exception for them would benefit society as a whole.
I never thought I'd say this before, but I hope Disney wins this copyright suit.
Yeah from the point of view of the law you're basically saying:
"well we've gotten so good at hoovering up random crap we can't tell if any of it is stolen", which, yeah ok. Doesn't matter? It means you've admitted you know you're stealing and it can be proved you're stealing. The fact you came up with a new way to collect loose crap doesn't change that fact, nor does it make it legal to do so.
I think there's a difference between having an artistic capability (it could generate an Iron Man image) and actually doing it. If the buyer explicitly names the character in the prompt, maybe it's their fault for asking for such a thing?
If it's generating such images for a generic prompt, that's quite another thing.
A simple Google image search looking for an existing thing will pull AI generated images without anyone inputting anything
For example, I saw a logo/initials as a sticker on someone's car that was KHK but the H looked very much "K" like due to the design (a slanted cross bar for example). So I tried to Google (like I've done in the past) for "KHK logos" "KHK initials looks like KKK" and a number of other iterations. The image search was full of helpful and as far as I can tell newly generated logos for me for the KKK.
I played with my settings but never found the exact thing I'm looking for. If I removed AI images I only got pics related to a Qatari (I think) company with the same name. So while maybe they're keyword blocking this from Disney itself I haven't tested, it's actually harder not to get forcefed AI images than not.
Sometimes the big companies are just plain at fault.
I'm a little unclear about what you saw. Are you saying that someone out there is flooding Google Image search with AI slop?
I think the issue is more with training data than output. Unauthorized use is still unauthorized, and copyright sucks with how restrictive it is in many countries.