Hum… I like the way you’re thinking.
If the SRS system had a way to memorize which words one knows, it could display only sentences where the user does know everything. I know that @neicul has been trying to figure out a way to do the known words counting in an efficient way on Kitsun.
This could probably not be a bad thing though, depending on how the whole thing is built 
I personally keep it simple: I read/watch something, see a word I’m interesting in learning, make a card out of it (post-exposure).
I do think that the best approach is the way that @Ncastaneda does: getting the content straight from the source (movie, TV series, etc) and turn it into a word/sentence card (?). Idk, tagging them as they might be interested in discussing it.