@Gargron it's a pretty good idea in my opinion. Maybe with many Keyframes or just 80% of the image reconstructed by machine learning it can be non-distinguishable even with environment changing fast etc.
Just goes to show what could be possible with slow connections, mesh networks etc.