AI art again. A lawsuit here, a new path over there

[KeegansCreation]

In the AI art world (which a lot of designers are dabbling in, myself included) two important things are happening in parallel:

1)Getty Images is suing Stable Diffusion and Midjourney (which is built on Stable Diffusion) on the claim that using their copyrighted images to train the model is copyright infringement. It is important to note that they do not make the claim that the end images infringe copyright, but rather that the training itself is the infringement. It is also important to note that they are not suing Open AI or DALL-E2 (which is built on Open AI).

2)Shutterstock has partnered with DALL-E2 such that DALL-E2 could train on their images via a license. Shutterstock also now has AI capabilities on their website. When I first looked their website, they had a portal to DALL-E2. Now they are just saying it is their own generator (which maybe it is or maybe it is still a portal). They also have a thing about their partnership with Open AI and DALL-E2 and how you can generate images that are guaranteed to be ethically sourced. This partnership does explain why DALL-E2 images tend to look more photo-ish (of course they also have the entirety of public domain art history on there, just like Midjourney).

I discovered this while trying to find out why Getty didn't sue Open AI/DALL-E2 (maybe the partnership with Shutterstock meant that they didn't scrape Getty?). I was also trying to figure out why Shutterstock wasn't making a suit of its' own.

The end result of this suit is probably many months away. DALL-E2 is very affordable and I hope it stays that way and this doesn't morph into having to pay Shutterstock to be able to use it. Fingers crossed.

DALL-E2 and Midjourney are currently neck and neck in the AI art race. I have no idea who will win this suit. I've read everything I could find and all the lawyer pundits seem to have no idea either.

If Getty wins, Midjourney will at best have to re-train either with no Getty images or with licensed Getty images or with Shutterstock images if that deal wasn't exclusive to DALL-E2. They may have to pay damages. Retraining will take months, allowing DALL-E2 to pull ahead.

If Midjourney wins, things continue on as they are now.

Who wins no matter what? Us, because DALL-E2 is in the clear no matter who wins or loses. DALL-E2 may be the new path per my title. Also Shutterstock because they see the writing on the wall and predict that AI is the future of stock and they just got in on the ground floor.

[PenguinPower · Accepted solution]

I think this is a bit silly… how do humans train themselves? By looking at and learning from what has come before them. As long as the AI isn’t doing things that are recognizably copies I don’t think they should have a case…

[Cat]

I agree that this is a bit silly. If anyone should be sued for copyright infringement it's the image search engines that basically scrape each and every image off of the web and then present it in their own format so people can easily get the image without ever visiting the page where it's posted - that basically ruined my entire former business because suddenly nobody had to go to my website to get my images, Google would just grab my images, bypass me entirely (and the ads which were my only source of revenue) and hand people my content!

[KeegansCreation]

Somebody actually did sue Google for that- an 'adult entertainment' company. But the court ruled in favor of Google.

[PenguinPower]

😂I can understand that... all they have to sell is the privilege of looking.. 

[Cat]

wow! Well it's very disheartening that the court found in favor of Google.

[KeegansCreation]

That's the linchpin. Does training constitute infringement, as Getty claims? I think this is actually a very important case since it has implications beyond AI art. ChatGPT was trained the same way, except writing rather than visual. But it scraped copyrighted written works for training. It hasn't been sued. Maybe people are hanging back waiting to see what precedent this case sets. And whichever precedent it sets is going to affect all types of AI going forward.

I agree that it shouldn't be infringing given that none of the outputs infringe and it does the same thing humans do just much faster. But the lawyer pundits aren't so sure it will be open and shut. There is already legal precedent for AI to not be treated the same way as humans in that AI works aren't copyrighted unless they are modified by a human. A human isn't infringing when studying other artists unless they make a literal copy of a specific work. So in the case of humans, only output can be infringing, not input. But will the court say that the same holds true for AI?

I am actually deeply interested in this case because I think it has implications that go far beyond whether or not Stability Diffusion has to retrain.

I wish I could be confident Getty will lose but the lawyer pundits seem completely unwilling to even guess so I can't be sure. Interesting days ahead in any case. 

[PenguinPower]

As you say who knows… there is definitely legal precedent that a non-human cannot hold a copyright (the monkey selfie case) but whether or not a non human can infringe… that is certainly new territory.
Getty already has my hackles up… they should be the last ones bold enough to do this, after they’ve been caught trying to bill the creator of an artwork for using her own photographs on her own web page after she specifically put into the images public domain… the situations are somewhat different, but still… they certainly take advantage of scraping whatever they can off of the web, so pot meet kettle

[Barbara]

The idea of AI learning from images on the internet suddenly reminded me that many serious oil painters have often honed their art by copying the paintings of masters. Growing up, we had just such a student's copy of "The Horse Fair" by Rosa Bonheur hanging over the fireplace. Absolutely gorgeous, but no more than half the size of the huge original. My father acquired it from someone who simply didn't want it anymore. Not only was it legal to copy the painting, but it was also legal to sell the copy.

Of course, the above is not the same as what's going on with AI, but there are certainly parallels.

Colorwash's Home