Today, images/audio/videos are sent over as inline bytes in a proto over the websocket. However, we could easily save them to a temporary location and serve them via HTTP. Then the proto would only need to contain the URL where that resource could be found in the Streamlit server.

To avoid having to manage cache invalidations, we could keep only 1 version of each image/audio/video and each user session. (Meaning the files get deleted with each rerun from its user session, or when the session disconnected)

This is not just an infra task but also impacts users. See: https://discuss.streamlit.io/t/st-audio-not-working/388/3