most of the criticism I've seen is that it's all trained on uncompensated stolen artwork. Much like how copilot is trained on GPL code, disregarding its license terms.
It's interesting to ask people who are concerned about the training data what they think of Adobe Firefly, which is strictly trained on correctly licensed data.
I'm under the impression that DALL-E itself used licensed data as well.
I find some people are comfortable with that, but others will switch to different concerns - which indicates to me that they're actually more offended by the idea of AI-generated art than the specific implementation details of how it was trained.
I then did works in the style of what I’d researched. I trained myself on works I didn’t own, and then produced my own.
I kind of see the AI training as similar work, just done programmatically vs physically.
Certainly a very interesting topic.
I can’t get my head around how far we’ve come on this in the last 6-12 months. From pretty awful outputs to works winning Photography awards. And prints of a dog called Queso you’d have paid a lot of money to an illustrator for.
That's not to say there aren't other ways you might use the direct image (e.g. collage or sampling in music) but you'll likely be careful with how it's used, how much you tweak it, and with attribution. I think the weird problem we're butting up against is that AFAIK you can't figure out post-facto what the "influence" is from the model output aside from looking at the input (which does commonly use names of artists).
I work on an AI image generator, so I really do think the tech is useful and cool, but I also think it's disingenuous (or more generously misinformed) to compare it to an artist studying great works or taking inspiration from others. These are computers inputting and outputting bits. Another human analog would be memorizing a politician's speech and using chunks of it in your own speech. We'd easily call that plagiarism, but if instead every 3 words were exactly the same? Hard to say... it's both more and less plagiarism.
Just how much do you need to process a sampled work before you need to get permission of the original artist? It seems to be in music that if the copyright holder can prove you sampled them, even if it's unrecognizable, then you're going to be on the hook for some royalties.
I don't think that's true.
My understanding is that any image generated by Stable Diffusion has been influenced by every single parameter of the model - so literally EVERY image in the training data has an impact on the final image.
How much of an impact is the thing that's influenced by the prompt.
One way to think about it: the Stable Diffusion model can be as small as 1.9GB (Web Stable Diffusion). It's trained on 2.3 billion images. That works out as 6.6 bits of data per image in the training set.
> My understanding is that any image generated by Stable Diffusion has been influenced by every single parameter of the model - so literally EVERY image in the training data has an impact on the final image.
That's pretty interesting. Need to dig into the math more (lazy applications dev).
If they truly got an appropriate license agreement for every image in the training set then I have no issues with that.
>I'm under the impression that DALL-E itself used licensed data as well.
DALL-E clearly used images they did not have a license for. Early on it was able to output convincing images of Pikachu and Homer Simpson. OpenAI certainly didn’t get licensing rights for those characters.
What ongoing litigation I'm aware of seeks to close that loophole and make fanart illegal, which would be a first step towards also preventing AI art.
Not sure if there would be trademark issues. But that would be the case regardless of how the image was created.
In practice? No, not really. But Pikachu is a copyrighted character and only those with license to do so are actually legally allowed to reproduce Pikachu in media.
Trademarks can come into play like you said, but even just base copyright allows for the ownership of characters such as Pikachu or Batman or whatever.
There is an irony, however, that many of the AI art haters tend to draw fanart of IP they don't own. And if Fair Use protections are weakened, their livelihood would be hurt far more than those of AI artists.
The Copilot case/lawsuit IMO is stronger because the associated code output is a) provably verbatim and b) often has explicit licensing and therefore intent on its usage.
Creating a derivative work of a copyrighted image requires permission from the copyright holder (i.e., a license) which many of these services do not have. So the real question is whether AI-generated "art" counts as a derivative work of the inputs, and we just don't know yet.
>b) often has explicit licensing and therefore intent on its usage
It doesn't matter. In the absence of a license, the default is "you can't use this." It's not "do whatever you want with it." Licenses grant (limited) permission to use; without one you have no permission (except fair use, etc. which are very specifically defined.)
That's why "fair use" is the key concept here. Under US copyright law "fair use" does not require a license. The argument is that AI generated imagery qualifies as "fair use" - that's what's about to be tested in the courts.
https://arstechnica.com/tech-policy/2023/04/stable-diffusion... is the best explanation I've seen of the legal situation as it stands.
Inspired/trained… I think these could be seen as the same.
If the person directly copied another work, that's a derivative work and requires a license. But if a person learned an abstract concept by studying art and later created art, it's not derivative.
Computers can't learn abstract concepts. What they can do is break down existing images and then numerically combine them to produce something else. The inputs are directly used in the outputs. It's literally derivative, whether or not the courts decide it's legally so.
Goalposts can be moved on whether it has "truly learned" the abstract concept, but at the very least neural networks have the ability to work with concepts to the extent that you can ask to make an image more "chaotic", "mysterious", "peaceful", "stylized", etc. and get meaningfully different results.
When a model like Stable Diffusion has 4.1GB of weights and was trained on 5 billion images, the primary impact of one particular training image may be very slightly adjusting what the model associates with "dramatic".
> If the person directly copied another work, that's a derivative work and requires a license
Not if it falls under Fair Use. Here's a fairly extreme example for just how much you can get away with while still (eventually) being ruled Fair Use: https://www.artnews.com/art-in-america/features/landmark-cop... - though I wouldn't recommend copying as much as Richard Prince did.
> The inputs are directly used in the outputs
Not "directly" - during generation, normal prompt to image models don't have access to existing images and cannot search the Internet.
I would say that abstract concepts is the only thing that computers can learn at the moment, at least until they are successfully embodied.
> It's literally derivative, whether or not the courts decide it's legally so.
To be a derivative work you should be able to at least identify the work it is a derivative of. While SD and friends can indeed generate obviously copyright infringing works (then again so can photoshop or a camera or even a paintbrush), for the vast majority of the output you can at best point out to the general direction of an author or a style.
It does not (in US law) if it falls within Fair Use, which is an exception to what would otherwise be the exclusive rights of copyright holders.
I think what you mean to say is that the argument is that both the models themselves and (in many cases) the output from the models, to the extent it might otherwise be a derivative work of one or more the input images, are transformative uses. [0]
[0] https://www.nolo.com/legal-encyclopedia/fair-use-what-transf...
But the term "derivative work" has specific legal meaning. For example wikipedia states: "[derivative work] includes major copyrightable elements of an original, previously created first work (the underlying work)".
I would say that while that might be the case for some AI creations, it is certainly not true for all (or even a majority) of them.