For educational purposes, this is being posted as a test of the real-time SRNN from the prior blog post.
Everything is pretty decent - and this version again only looks at a frame at a time and doesn't consider past or future frames for the super-resolution (thus vast room for improvement).
It has issues where it magnifies/interprets DCT artifacts as meaningful - but I suppose that's to be expected. Garbage in -> Garbage out. Ideally you would use this with uncompressed data.
In theory you could train a NN to remove the DCT artifacts before processing or maybe do an all-in-one - though that's probably a diminishing returns kind of scenario.
The original file incase the host re-encodes it in less than 4k...