I honestly think splitting communities in smaller and smaller subsections is detrimental. A look at how things worked during the formative years of the internet (specifically WWW) might be helpful: Allow all content that's not straight-up illegal, and leave it to the users to filter whatever they don't want to see. You as a user can block certain people/communities, or switch to the "Subscribed" view to only see what you like, just like how it was on Reddit. That's a better use of everyone's time than a) getting more and more granular with what's and what's not allowed in AI (!) art, and b) immediately resorting to the nuclear option, which is a community split, defederating another instance, or similar measures.
In general, we should avoid treating instances like specific subreddits, with very micromanaged rules and such. The instance is infrastructure, it should only deal with content insofar it is unhealthy for the network as a whole (banning bot instances, for example). Content on the other hand should be moderated by moderators on the community level (if you don't like what you see, just make a new one and stop subscribing to the old one).
This is better for network and community health, for the admins, and for proliferation of Lemmy in general.