I would argue (as the court did) that google's use is transformative because the end result "book search" is in a different marketplace from "books." The end result / output of these generative AI systems trained on stock media and art is..."stock media and art."
That's kind of what this whole article is about. Just training the systems in research is arguably fair use but creating the entire pipeline might not be and the "loophole" here is trying to claim no responsibility for the training at the center of it because that was technically done by a 3rd party (...funded by the final creator of the full entire pipeline.)
The court’s summary also mentions this aspect of differing marketplaces:
“… the revelations [i.e. the information served by Google Book Search] do not provide a significant market substitute for the protected aspects of the originals.”
This doesn’t apply to AI image generators which are clearly a “market substitute” for the protected originals used to train the system. For this reason I’d expect someone like Getty to want to revisit Authors Guild v Google sooner rather than later.
Can we first get an AI that's actually usefull as a Getty substitute? All I'm seeing posted is visually pleasing nonsense - as soon as I tried to use it for stuff like stock photo generator it's unusable (eg. key physical properties of the object would be off to the point where the object is useless, and in many cases it would look wrong even from a thumbnail).
The only thing I did see was designers cropping out the wrong part and filling in the blanks - I suppose it's competing with stock photos in that aspect.
Did you try inpainting to fix the wrong bits? From my very little experience, AI image generation is not (yet) a one click process and requires multiple iterations to get close to the desired result, i.e. it is still more a tool for a designer than a replacement for it.
AI image generators are clearly not a market substitute for images, they are a tool that can be used to create market substitutes, but not themselves one.
Depends on what the product is. For example with openAI Dalle-2 the product is very clearly the generated image. You even pay per image. Also this is kind of what this article is about. Arbitrarily separating the pieces in order to evade copyright.
That's kind of what this whole article is about. Just training the systems in research is arguably fair use but creating the entire pipeline might not be and the "loophole" here is trying to claim no responsibility for the training at the center of it because that was technically done by a 3rd party (...funded by the final creator of the full entire pipeline.)