Couldn’t images and videos just be loaded from the instance they were uploaded to instead of getting copied to each instance? It would work almost the same as uploading it to a file hoster but it would be a lot easier usability wise and illegal content would still only have to be deleted at a single point.
When you log into a new client, you should get a request on a client you were already logged into where you just compare if the emojis are the same and then it should just send over the decryption keys. Granted, it hasn’t always worked for me either but it does most of the time. And you’ll be able to see new messages either way.