Seems like a very lossy form of compression. Not to mention that hashes usually aren't reversible. Not a great stand in for the real thing, especially when the real thing might be a widely available jpeg. There's really something I don't understand about how an NFT, apart from cryptographic foundation, is different enough from, say, a UUID in a traditional database.
Also it would be funny if there was a (very highly unlikely) hash collision between two different valuable NFTs.
Whoever owns an NFT can prevent the pre-image from being lost by just personally storing it. If they wanted to ensure the pre-image is easily accessible to others, they could additionally host it on IPFS.
What if the NFT is for a physical object? (and one that may or may not exist any longer?)
I mean, clearly hashing wouldn't seem to apply there. But it still leaves the question of whether there's a substantial difference between a traditional database entry. I may be overlooking something because I don't see the problem solved by NFTs. I see the use for them as part of an artistic statement, but not a general use function.
Also it would be funny if there was a (very highly unlikely) hash collision between two different valuable NFTs.