Frankly, one of the big reasons why I like the Fediverse is that you don’t have to depend on the source keeping up their stuff - we get our own copies. That includes both the text (the posts) and the media (the images), and to me, that’s one third of the point of the entire thing. If I wanted an image that can only be seen at the dumpster cage that is Twitter, I’d go over there shrug and move on.

Of course that’s not always workable because storage is (despite everything) not cheap, medias grow large (I can eg.: understand saving pictures, but heaven forbid trying to save a video) and there’s still not a good way to deal with “”“problematic”“” storage. So, my recommendation and expectation would be something that functions like Option 2: Proxying Images. Basically, we download our own copy but only store it “while it matters”.

How to Supplement Options

Now, maybe some proposals I would lift to have their feasibility studied. Any combination of one or more of these could, if implemented, help in enhancing or even supplantting the chosen method for storing remote media. I personally see them more as a means to enhance Option 2.

• Only save copies of images (perpetual or proxy) that are below certain thresholds: file size, resolution, trusted hosts, etc… Sure, that still means you have to download every image at least once to evaluate it, but at least we get some sort of automated guarantee that for “easy” stuff, we won’t be filing more connections than necessary. The big win of this option is that lemm.ee is not paying for the larger resource cost of fetching larger images after their post is made. I’m guessing the main drawback would be the bikeshedding required to decide which images get saved locally.
• Related option to the above: only save (perpetual) thumbnails or smallened versions of images, never the real ones. Would pretty much instantly cover the case of eg.: most memes. The big drawback I can see to this is managing those images for deletions would be harder unless a Lemmy instance can keep a searchable map of hashes from each image to their thumbnail and vice versa.
• Require that any image is linked from an imagehost or filehost that is trustable for durability. Big drawback: notoriously more effort (and potential loss of privacy) for users means this actually disincentivizes posting rich content.
• Pool up resources and get some cooperative work going with some other instance(s) to set up a shared proxy agent that downloads the images for us, so that lemm.ee doesn’t have to host the images but can have the clients fetch them without sacrificing privacy. I feel this one incentivizes posting rich content because you can get some level of assurance that it’ll remain available and be “cheap” to access from across the Fediverse, but requires more instances to chip in.
• Images? Pfff. Text is where it’s at.