Copyrightability of GenAI output
Note: I'm not a lawyer so this is to be taken as an opinion on a topic that's far from being settled.
So far the way that I see and often interpret copyright around GenAI
(such as with Diffusion Models and Large Language Models) is that the generation
itself might not be copyrightable.
Specially when the user input itself, isn't.
Except, this doesn't means all the output it can make is entirely non-copyrightable, for example let's say automatically scanning a book isn't a copyrightable act, it doesn't automatically means it would also strip out the copyrights status of the books getting scanned.
And if you think something short is somehow fine legally,
see for example the cases around Google News.
Which started with aggregating newspapers into title + picture +
quote without their prior permission.
Newspapers and news agencies then sued for copyright violation, and at least some have won their lawsuits: