I think replicating DOM & sending diffs instead of video should 1) saves resources on encoding for server 2) saves decoding on client 3) send much lower data 4) feels much better since instant scrolling/hover (something like https://www.rrweb.io/)
I'm indie/solo making Linkkraft browser (to make a living from it). Browser to be effective researcher & collector. It visualizes your steps as tree and makes html snapshot for your each step (even steps in SPAs like twitter). https://arestov.github.io/linkkraft-notes/comparing/linkkraf... https://arestov.github.io/linkkraft-notes/trails-tree-plus-o...
As side effect of snapshots you can confidently unload documents & save CPU/memory.
Looking forward to grow that snapshotting part into DOM streaming. So you can run "browser server" on your own PC, while having laptop fast & cool. With almost 0 delays and without worrying about privacy.