We've been wanting to collab with a client dev on this for a while so this is great to see, happy to talk through. Another option to consider instead of a NIP is standardizing on the structure of an event for media. One big reason for this is it could be a lot more flexible for a variety of media types by using the MIME standard as a descriptor. We wrote this up close to a year ago but I think it's worth a look (it's also what we use to feed Wavman): https://github.com/wavlake/nom-spec