Images is not my domain, but it would make sense to ask photographers or other people in that realm what the type of features they can think off that leverage kind20 tags data which could be implemented in kind20 clients to suit their needs.
I disagree with "keep nostr client implementation simple" to begin with, in the sense that you could just choose to opt for a simple (half-assed, feature poor) implementation of kind, rather than encouraging a shitty practice that undermines feature rich specific clients.