It might be okay if there was a way for relays to reject content based in its own criteria. If a small nsfw model could be run on the content before rejecting or purging the images/videos things it would be a little more manageable.

Still, need relays that dont implement that nip just in case though.