Those images look nothing alike unless you stop looking beyond the contrasted regions… Which, fair enough, could indicate someone taking the outline of the original, but you hardly need AI to do that (Tracing is a thing that has existed for a while), and it’s certainly something human artists do as well both as practice, but also just as artistic reinterpretation (Re-using existing elements in different, transformative ways).
It’s hard to argue the contrast of an image would be subjective enough to be someone’s ownership, whether by copyright or by layman’s judgement. It easily meets the burden of significant enough transformation.
It’s easy to see why, because nobody would confuse it with the original. Assuming the original is the right, it looks way better and more coherent. If this person wanted to just steal from this Arcipello, they’re doing a pretty bad job.
EDIT: And I doubt anyone denies the existence of thieves, whether using AI or not. But this assertion that one piece can somehow make sweeping judgements about multi-faceted tech by this point at least hundreds of thousands if not millions of people are using, from hobbyist tinkerers to technical artists, is ridiculous.
You are speaking bollocks, there are already many lawsuits by artists against the so called Ai engines, there are boundaries on how much you can copy from a specific artwork, logo, design or whatever, for example if you take the coca cola logo and slightly change it even if it doesn’t say coca cola you will still face the laws of copyright infringement, nobody denies the existence of thieves, so that’s why people do whatever they can to protect their work
Lawsuits, yes. But a lawsuit is not by default won, it is a assertion for the court to rule on. And so far regarding AI, none have been won. And yes, there are boundaries on when work turns into copyright infringement, but those have specific criteria, and regions of contrast do not suffice by any measure. Yes, even parts of the Coca Cola logo can be reinterpreted without infringing. Why do you think so many off brands skirt as close as possible to it without infringing?
AI can absolutely produce copyrighted content if it’s prompted to. Name drop an artist in Midjourney and you will be able to prompt their style - see this list of artists and prompted images. So you can just tweak the settings a bit to heavily weight their name, generally describe the composition of the work you’re looking to approximate, and you can absolutely produce something close to their original works.
The image is wrong because the original artwork is not stolen. It is part of a dataset by LAION (or another similar dataset, basically a text-image pair where the image is linked at its original source). To train the imagegen, its company had to download a temporary copy, which is exempt from infringement by copyright law. There is no original artwork somewhere in a database accessible by Midjourney, just the numerical relationship generated by the image-text pair it learned from.
On the other hand, AI can obviously produce content in violation of copyright - like here. But that’s specifically being prompted by the user. You can see other examples of this with Grok generating Mickey Mouse and Simpsons characters. As of right now, copyright violations are the legal responsibility of the users generating the content - not the AI itself.
I think you meant to respond to someone else, as I pretty much agree(d) with everything you’re saying and have not claimed otherwise. In fact in my very post I did say in more layman terms it was very likely this person used img2img or controlnet to copy the layout of the image, I think it’s less likely they got something this similar unguided, although it’s possible depending on the model or by somehow locking the prompt onto the original work.
But the one point I do disagree with is that this is a violation of copyright, as I explained before. For it to be a violation it would need to look substantially more similar to the original, the one consistent element between the two is the rough layout of the image (the contrasted areas), for the rest most of the content is very different. You notice the similarity of the contrasted area much more easily by it being sized down so much.
I hope you understand, as you seem to be more knowledgeable than the people that downvoted without leaving a comment, but you are allowed to use ideas and concepts from others without infringing on their work, as without it the creative industry literally couldn’t function. And yes, this is the responsibility on anyone using these models to avoid.
This person skirts too close in my eyes by pretty much 1:1 copying the layout, but it’s almost certainly still fine as again, a human doing this with an existing piece of work would also be (eg. the many replica’s / traces of the Mona Lisa).
Hell, if you take a look at the image in this very lemmy post, which was almost certainly taken from someone else, it has a much better case of copyright infringement, since it has the same layout, nearly identical people in the boxes, general message and concepts.
But in the end, copyright is different per jurisdiction and sometimes even between judges. Perhaps there is a case somewhere. It’s just (in my opinion) very unlikely to succeed based on the limited elements that are substantially similar.
EDIT: Added the section about the Mona Lisa replica’s for further clarification.
Hm yeah on second look the images aren’t as comparable as I expected. I just saw the general composition in the thumbnails and assumed more similarity. I do think they probably prompted the original artist in the generated work, though, which kind of led to my thoughts in my op.
Yeah that’s also fair enough conclusion, I think it’s a bit too convenient the rest of the image looks a lot worse (Much more clear signs of botched AI generation) while the layout remains pretty much exactly the same, which to me looks like selective generation.
“Theft” only applies to the poor. Rich assholes and their megacorps will pay judges to tell you so
Those images look nothing alike unless you stop looking beyond the contrasted regions… Which, fair enough, could indicate someone taking the outline of the original, but you hardly need AI to do that (Tracing is a thing that has existed for a while), and it’s certainly something human artists do as well both as practice, but also just as artistic reinterpretation (Re-using existing elements in different, transformative ways).
It’s hard to argue the contrast of an image would be subjective enough to be someone’s ownership, whether by copyright or by layman’s judgement. It easily meets the burden of significant enough transformation.
It’s easy to see why, because nobody would confuse it with the original. Assuming the original is the right, it looks way better and more coherent. If this person wanted to just steal from this Arcipello, they’re doing a pretty bad job.
EDIT: And I doubt anyone denies the existence of thieves, whether using AI or not. But this assertion that one piece can somehow make sweeping judgements about multi-faceted tech by this point at least hundreds of thousands if not millions of people are using, from hobbyist tinkerers to technical artists, is ridiculous.
You are speaking bollocks, there are already many lawsuits by artists against the so called Ai engines, there are boundaries on how much you can copy from a specific artwork, logo, design or whatever, for example if you take the coca cola logo and slightly change it even if it doesn’t say coca cola you will still face the laws of copyright infringement, nobody denies the existence of thieves, so that’s why people do whatever they can to protect their work
Lawsuits, yes. But a lawsuit is not by default won, it is a assertion for the court to rule on. And so far regarding AI, none have been won. And yes, there are boundaries on when work turns into copyright infringement, but those have specific criteria, and regions of contrast do not suffice by any measure. Yes, even parts of the Coca Cola logo can be reinterpreted without infringing. Why do you think so many off brands skirt as close as possible to it without infringing?
They don’t! And most of those lawsuits are still in process
Thats what I said, yes.
AI can absolutely produce copyrighted content if it’s prompted to. Name drop an artist in Midjourney and you will be able to prompt their style - see this list of artists and prompted images. So you can just tweak the settings a bit to heavily weight their name, generally describe the composition of the work you’re looking to approximate, and you can absolutely produce something close to their original works.
The image is wrong because the original artwork is not stolen. It is part of a dataset by LAION (or another similar dataset, basically a text-image pair where the image is linked at its original source). To train the imagegen, its company had to download a temporary copy, which is exempt from infringement by copyright law. There is no original artwork somewhere in a database accessible by Midjourney, just the numerical relationship generated by the image-text pair it learned from.
On the other hand, AI can obviously produce content in violation of copyright - like here. But that’s specifically being prompted by the user. You can see other examples of this with Grok generating Mickey Mouse and Simpsons characters. As of right now, copyright violations are the legal responsibility of the users generating the content - not the AI itself.
I think you meant to respond to someone else, as I pretty much agree(d) with everything you’re saying and have not claimed otherwise. In fact in my very post I did say in more layman terms it was very likely this person used img2img or controlnet to copy the layout of the image, I think it’s less likely they got something this similar unguided, although it’s possible depending on the model or by somehow locking the prompt onto the original work.
But the one point I do disagree with is that this is a violation of copyright, as I explained before. For it to be a violation it would need to look substantially more similar to the original, the one consistent element between the two is the rough layout of the image (the contrasted areas), for the rest most of the content is very different. You notice the similarity of the contrasted area much more easily by it being sized down so much.
I hope you understand, as you seem to be more knowledgeable than the people that downvoted without leaving a comment, but you are allowed to use ideas and concepts from others without infringing on their work, as without it the creative industry literally couldn’t function. And yes, this is the responsibility on anyone using these models to avoid.
This person skirts too close in my eyes by pretty much 1:1 copying the layout, but it’s almost certainly still fine as again, a human doing this with an existing piece of work would also be (eg. the many replica’s / traces of the Mona Lisa).
Hell, if you take a look at the image in this very lemmy post, which was almost certainly taken from someone else, it has a much better case of copyright infringement, since it has the same layout, nearly identical people in the boxes, general message and concepts.
But in the end, copyright is different per jurisdiction and sometimes even between judges. Perhaps there is a case somewhere. It’s just (in my opinion) very unlikely to succeed based on the limited elements that are substantially similar.
EDIT: Added the section about the Mona Lisa replica’s for further clarification.
Hm yeah on second look the images aren’t as comparable as I expected. I just saw the general composition in the thumbnails and assumed more similarity. I do think they probably prompted the original artist in the generated work, though, which kind of led to my thoughts in my op.
Yeah that’s also fair enough conclusion, I think it’s a bit too convenient the rest of the image looks a lot worse (Much more clear signs of botched AI generation) while the layout remains pretty much exactly the same, which to me looks like selective generation.