If there were a system that could uniquely identify each and every single file that is uploaded to it and then on the fly cross compare that file with previous uploads and identify when the file is a duplicate, that would be very handy.
That exists. And we could probably even make something that recognizes small changes like added text on the image and save the data accordingly, probably using some kind of neural network as you later mentioned. The issue is just that that would be extremely computationally expensive, so companies usually choose to just buy a shitload of HDDs. Maybe something like this might be more feasible with quantum computing at some point.
Something like NFTs could be used to accomplish that, if it were fully fleshed out and developed.
What makes you think that? An NFT is basically just an ID card with extra steps. I don’t see how NFTs relate to the problem you described at all. Just because most NFTs the public heard about were somehow tied to images or videos, doesn’t mean that NFTs are a technology for detecting duplicate images/data in general.