Alt text should describe what you’re communicating, not necessarily every detail.

Eh, maybe? Most of the transcriptions I've seen, particularly the ones from the explicitly accessibility-focused now-defunct Transcribers of Reddit (the founder of which I was in close communication with, as a co-mod of an unrelated subreddit) try to simply describe what you're seeing. In enough detail that those in the know might be able to make the same conclusions as a sighted person, but without providing details that aren't self-evident from the original source. So like, if it was a picture of a known character, it might say that, but if it just vaguely looks like an amalgamation of a couple of fictional characters, you'd be unlikely to see that specified, especially if they're relatively obscure and less likely to be familiar to most of the audience.