Interesting. I wonder what the copyright office would think of the recent ControlNet, which allows you to turn, among other things, crude sketches into pictures or photos. Would that still not...
The USCO said that since Midjourney produces an image first by "randomly generating noise," the process wasn't something that could be directly controlled by humans. Text prompts fed into the software don't guarantee it will create a particular image – therefore it is not a tool like Photoshop or GIMP.
Interesting. I wonder what the copyright office would think of the recent ControlNet, which allows you to turn, among other things, crude sketches into pictures or photos. Would that still not count as "directly controlled by humans" because the details are filled in by the model? How detailed would an input sketch have to be to count? To me, it seems like they've taken themselves onto a quite slippery slope here.
But that's exactly what creates the slippery slope. For example, from what I understand, Photoshop has a so-called brush engine that can randomly jitter the brush strokes to create various...
Yes, because the artificial intelligence system is making decisions without human input. Yes, the generated work is overall created by an input given by a human (presumably), but there are individual steps in the generation process that are decided specifically by the system itself. In other words, the human is not the only artist of the work, and the other artist is not human.
But that's exactly what creates the slippery slope. For example, from what I understand, Photoshop has a so-called brush engine that can randomly jitter the brush strokes to create various artistic effects using random numbers. It would be preposterous to deny copyright for an image created using that feature; the question I was posing is: where does the human artist end and the machine artist begin (assuming there is a machine artist that is not just a tool)?
For example, I can't scribble this shitty rendition of a dog and then go claim it as a copyrighted work that I have sole permission to use. I could certainly try, but even if I was granted a copyright to that work, its extreme triviality would make enforcing my copyright on any works derived from it almost impossible.
Again, that's exactly what I meant with my slippery slope argument. Approximately anyone can scribble a shitty rendition of a dog, and I'd agree that it should probably not be protected. But what if the input is an actually artistical line drawing and the system colorizes or enhances it? Is there enough that is "directly controlled by a human" to invalidate the current claim of "it's just random noise" that resulted in the comic not being recognized?
I think thats where it falls under either "trivial enough edits its more like a tool" or "If you want to claim copyright, you must own the copyrights to the training data." So if you take a bunch...
But what if the input is an actually artistical line drawing and the system colorizes or enhances it?
I think thats where it falls under either "trivial enough edits its more like a tool" or "If you want to claim copyright, you must own the copyrights to the training data."
So if you take a bunch of public domain art to train an AI model for yourself, then use that to do the work, there may be a claim in there. But if you train it on a bunch of stuff, like say from cartoons you don't have copyright to, that would void it. In the same way that copying an essay and running it through a thesaurus is still plagerism.
How certain are you of that? Absolutely genuine question, because my understanding was that you can totally do so: you can't claim the general field of "shitty renditions of a dog", but that...
For example, I can't scribble this shitty rendition of a dog and then go claim it as a copyrighted work that I have sole permission to use.
How certain are you of that? Absolutely genuine question, because my understanding was that you can totally do so: you can't claim the general field of "shitty renditions of a dog", but that precise expression of the idea is indeed the author's. As far as I'm aware you generally don't need to be granted a copyright at all - the work becomes copyrighted automatically as a function of you having created it - but formal registration can be helpful in proving the facts of a work's authorship and defending if litigation does happen. Which in turn would explain why the author of a work expected to be contentious like this would proactively register it ASAP.
Of course, if that image became a logo for my business, or some other "concerned component" of my intellectual property, then my ability to receive (and hold) copyright over it could improve.
I would have expected that to move into trademark rather than copyright territory?
As for the rest, I'd be interested to know how your thoughts play against the fact that shitty snapshot photos are considered the copyrighted property of the people who took them? I specify "shitty snapshot" because I absolutely understand the level of effort that can go into photography, but the effort or lack thereof doesn't have a bearing on whether the final photo is copyrighted.
I can point my phone out the window and get a poorly framed image of a tree and some cars in less than a second. Less time than it takes to type even a basic prompt - and a lot less time than it takes to refine a good prompt and guide it into worthwhile output - with the camera doing all of the meaningful work, but I still own that photo.
None of this is to say that I think IP in ML generated work is a simple topic, or even that I feel as though I have a strong enough grasp to have a specific answer. I haven't even touched on the training datasets or the competing interests at play there, for a start. I also consider current IP law to be pretty seriously broken in multiple ways. I just also think that a blanket stance like this from the copyright office seems to fly in the face of quite a few existing day-to-day examples, and I do find that concerning.
Yes, you could. But odds of actually being able to enforce it via a lawsuit would be slim. There's entire swaths of legal precident about needing to be sufficiently unique and noteworthy....
How certain are you of that? Absolutely genuine question, because my understanding was that you can totally do so.
Yes, you could. But odds of actually being able to enforce it via a lawsuit would be slim.
There's entire swaths of legal precident about needing to be sufficiently unique and noteworthy. Otherwise it would be impossible to copyright music in a remotely sane way (tho this is already somewhat true). Fair use like excerpts rides on the fact that taking a small excerpt of a larger work (for say a critique) doesn't infringe the copyright.
In that same vein, trivial code can't be enforcibly copyrightten (like an if-else block with some basic string comparisons).
If your drawing is comparable to anything a 4 year old that can hold a pencil can do, you probably can't get an enforcible copyright. Your lawsuit would be dismissed in short order.
The difference between what you've said here and what the USCO have said does an excellent job of highlighting why their take makes me uncomfortable. It's a complex question of what can be...
The difference between what you've said here and what the USCO have said does an excellent job of highlighting why their take makes me uncomfortable. It's a complex question of what can be enforced, built on precedent and test cases and the specific nuances and circumstances of the exact work in question and who's using it and for what. Even if we don't agree on the specifics of how a particular case should or would go, we're both approaching the debate as a whole from that same mindset.
Compare that to "Based on the record before it, the Office concludes that the images generated by
Midjourney contained within the Work are not original works of authorship protected by
copyright.". That's a pretty significant, sweeping statement on an entire class of works - and it's not just about enforceability, it's about whether copyright applies to them at all, ever. They do at least leave the door open for other tools a little later: "It is possible that other AI
offerings that can generate expressive material operate differently than Midjourney does.", but even just covering Midjourney outputs as a whole seems like a large and fairly risky precedent to set without significantly more robust consideration than we're seeing here.
I'm not a lawyer but this part doesn't sound right: The logo of a business is a trademark and I don't think trademark law has anything to do with copyright law? Also I haven't heard of a...
I'm not a lawyer but this part doesn't sound right:
Of course, if that image became a logo for my business, or some other "concerned component" of my intellectual property, then my ability to receive (and hold) copyright over it could improve.
The logo of a business is a trademark and I don't think trademark law has anything to do with copyright law? Also I haven't heard of a "concerned component" before. Are we even talking about the same legal system?
I think that when answering questions about law, it would be better to find experts who say whatever it is you remember and quote them, so people know where you got it from.
Another interesting example would be AI upscaling; is an upscaled image no longer protected since those gaps were filled in partially non-deterministicly?
Another interesting example would be AI upscaling; is an upscaled image no longer protected since those gaps were filled in partially non-deterministicly?
Interesting. I wonder what the copyright office would think of the recent ControlNet, which allows you to turn, among other things, crude sketches into pictures or photos. Would that still not count as "directly controlled by humans" because the details are filled in by the model? How detailed would an input sketch have to be to count? To me, it seems like they've taken themselves onto a quite slippery slope here.
But that's exactly what creates the slippery slope. For example, from what I understand, Photoshop has a so-called brush engine that can randomly jitter the brush strokes to create various artistic effects using random numbers. It would be preposterous to deny copyright for an image created using that feature; the question I was posing is: where does the human artist end and the machine artist begin (assuming there is a machine artist that is not just a tool)?
Again, that's exactly what I meant with my slippery slope argument. Approximately anyone can scribble a shitty rendition of a dog, and I'd agree that it should probably not be protected. But what if the input is an actually artistical line drawing and the system colorizes or enhances it? Is there enough that is "directly controlled by a human" to invalidate the current claim of "it's just random noise" that resulted in the comic not being recognized?
I think thats where it falls under either "trivial enough edits its more like a tool" or "If you want to claim copyright, you must own the copyrights to the training data."
So if you take a bunch of public domain art to train an AI model for yourself, then use that to do the work, there may be a claim in there. But if you train it on a bunch of stuff, like say from cartoons you don't have copyright to, that would void it. In the same way that copying an essay and running it through a thesaurus is still plagerism.
This is murky, but then, so is all copyright law.
How certain are you of that? Absolutely genuine question, because my understanding was that you can totally do so: you can't claim the general field of "shitty renditions of a dog", but that precise expression of the idea is indeed the author's. As far as I'm aware you generally don't need to be granted a copyright at all - the work becomes copyrighted automatically as a function of you having created it - but formal registration can be helpful in proving the facts of a work's authorship and defending if litigation does happen. Which in turn would explain why the author of a work expected to be contentious like this would proactively register it ASAP.
I would have expected that to move into trademark rather than copyright territory?
As for the rest, I'd be interested to know how your thoughts play against the fact that shitty snapshot photos are considered the copyrighted property of the people who took them? I specify "shitty snapshot" because I absolutely understand the level of effort that can go into photography, but the effort or lack thereof doesn't have a bearing on whether the final photo is copyrighted.
I can point my phone out the window and get a poorly framed image of a tree and some cars in less than a second. Less time than it takes to type even a basic prompt - and a lot less time than it takes to refine a good prompt and guide it into worthwhile output - with the camera doing all of the meaningful work, but I still own that photo.
None of this is to say that I think IP in ML generated work is a simple topic, or even that I feel as though I have a strong enough grasp to have a specific answer. I haven't even touched on the training datasets or the competing interests at play there, for a start. I also consider current IP law to be pretty seriously broken in multiple ways. I just also think that a blanket stance like this from the copyright office seems to fly in the face of quite a few existing day-to-day examples, and I do find that concerning.
Yes, you could. But odds of actually being able to enforce it via a lawsuit would be slim.
There's entire swaths of legal precident about needing to be sufficiently unique and noteworthy. Otherwise it would be impossible to copyright music in a remotely sane way (tho this is already somewhat true). Fair use like excerpts rides on the fact that taking a small excerpt of a larger work (for say a critique) doesn't infringe the copyright.
In that same vein, trivial code can't be enforcibly copyrightten (like an if-else block with some basic string comparisons).
If your drawing is comparable to anything a 4 year old that can hold a pencil can do, you probably can't get an enforcible copyright. Your lawsuit would be dismissed in short order.
The difference between what you've said here and what the USCO have said does an excellent job of highlighting why their take makes me uncomfortable. It's a complex question of what can be enforced, built on precedent and test cases and the specific nuances and circumstances of the exact work in question and who's using it and for what. Even if we don't agree on the specifics of how a particular case should or would go, we're both approaching the debate as a whole from that same mindset.
Compare that to "Based on the record before it, the Office concludes that the images generated by
Midjourney contained within the Work are not original works of authorship protected by
copyright.". That's a pretty significant, sweeping statement on an entire class of works - and it's not just about enforceability, it's about whether copyright applies to them at all, ever. They do at least leave the door open for other tools a little later: "It is possible that other AI
offerings that can generate expressive material operate differently than Midjourney does.", but even just covering Midjourney outputs as a whole seems like a large and fairly risky precedent to set without significantly more robust consideration than we're seeing here.
I'm not a lawyer but this part doesn't sound right:
The logo of a business is a trademark and I don't think trademark law has anything to do with copyright law? Also I haven't heard of a "concerned component" before. Are we even talking about the same legal system?
I think that when answering questions about law, it would be better to find experts who say whatever it is you remember and quote them, so people know where you got it from.
Another interesting example would be AI upscaling; is an upscaled image no longer protected since those gaps were filled in partially non-deterministicly?
but AI is just pasting together bits of art that humans drew from its training set
An excellent counter to that argument, from a related topic a few months back