

Another option could be to have the originating Mastodon server provide the preview image in content-adressable way, and have a standard allowing websites to provide the same: if the hash of the file(s) provided by the first / a Mastodon server to another matches the one from the target website, no need to download from the target website. Only if they differ then you’d actual download the data from the target website.
Each Mastodon server from the federation could quickly decide whether to trust that preview data from the peer it gets the rest of the post from or not.
This is basically taking ideas from IPFS but applying it to just the preview data that ends up being downloaded over and over again. You could also make that part a blob of data stored directly in IPFS. But IPFS was pretty slow to resolve content, when I last tested.




There are ways around that. See my other comment replying to the same parent as your comment.