if you want a very rough example of this, take a look at #[7]​ - a micro app for sharing and remixing audio tracks, of which media is core.

the media protocol is experimental and a WIP but we’re closer to your ideas than b64’d binary in json:

- we built a custom nostr relay that exposes a couple of non-websocket HTTP endpoints

- we upload media using multipart-form data. this is simple for clients to implement and chunks data on the way up

- since our relay powers a music streaming app we transcode the audio uploads for streaming (HLS). the relay currently doubles as the streaming fileserver

- our webapp emits a note on successful upload which includes the media urls

- when listening to media in stemstr, each note is rendered with a custom audio player that buffers in the audio as needed (streaming)

again, this is still a WIP and has lots of room for improvement but I wanted to add this data point to the discussion as this is a real world example of streaming media on nostr.