What does HackerNews think of Anime4K?

A High-Quality Real Time Upscaler for Anime Video

Language: Jupyter Notebook

Personally, it's not really any better than mpv filters like NIS [0] (Nvidia Image Scaling, which does not use AI but is somehow more sharp than these filters, likely because Nvidia used what they learned from AI upscaling to create a better non-AI upscaler; AMD's FidelityFX Super Resolution is also available as an mpv filter on the same page at [0]) as well as Anime4k [1] (which does use AI but only in the training phase to create the initial glsl file which then does not use AI inferencing but plain old image filters).

Combine mpv with its yt-dlp [2] and SponsorBlock support [3] and you have a competent way to upscale your own YouTube (and other types of [4]) videos.

If you really want to upscale with AI (albeit not at runtime), I've found Topaz Video AI to be the best so far, much sharper than other solutions on the market.

[0] https://gist.github.com/agyild

[1] https://github.com/bloc97/Anime4K

[2] https://www.funkyspacemonkey.com/replace-youtube-dl-with-yt-...

[3] https://github.com/po5/mpv_sponsorblock

[4] https://github.com/yt-dlp/yt-dlp/blob/master/supportedsites....

Even digital video created with low-resolution cameras will eventually be viewable in 4K, 8K and beyond because AI upscaling technology is just so great. See Anime4K for examples.

https://github.com/bloc97/Anime4K

Yes it "guesses" what the missing information is and populates it. Similar approaches will work for live action media too.

And successive video frames encode so much extra resolution that a single frame doesn't capture, so there's a lot of information to pull from.

There is also Anime4K [1]. It doesn't use machine learning though.

[1]: https://github.com/bloc97/Anime4K