Comment by Mithaldu

It should also be possible to train it on itself to improve moving scenes by using the motion itself as temporal super-sampling, just like the human eye does.