The differentiator would be semantic classification of the content itself, like "videos with your cat", "photos of your mountains" etc. I know Google Photos does it, but I haven't used it that much, not sure if b/c UX is bad or b/c maybe it doesn't solve an urgent need

also, the lowest hanging fruit might be helping me delete junk in my phone. but I understand this would likely include more than just metadata inspection and LLM-based sorting.

my main use case is deleting photos that are obviously useless, top of the line are memes I've sent people 1 month ago, duplicate photos, obviously wrong shots etc.

but might be just as hard as making a spam filter. maybe tarpit idea? I'd try it if I'd have the time, but it likely requires training, thus significant upfront costs