1j01 / rezzy-zoom-and-enhance Goto Github PK

Currently I use waifu2x-converter-cpp for superresolution, on a server which lives outside of the browser extension, and communicates with the web page over socket.io.
The biggest issue with this browser extension is that it's hard to set up. I would have published it already if it didn't require a separate install from the browser extension.

I just did a proof of concept for using Anime4K.js on a branch. There's performance problems (blocking the page), but that's easy to resolve, just need the computation to be in a worker (or the extension's background page). Cross-origin images are no longer handled, but that's solvable as well. The real problem is the quality of the output. Text looks very blurry/ugly. I wonder if the upstream project Anime4K has improvements that the JS port lacks, or if text is just a bad case for Anime4K.

There are other shaders and algorithms to compare.

• Hidden links should be de-prioritized / excluded
• Prioritize links in navigation
  • nav, .paginate, .pagination, .pages, .nav, .navigation e.g. (but much looser)
  • Prioritize links nearby in hierarchy to a paired next/prev link (could be an unreliable factor if links are hidden at the first / last page)
• Penalize links in a comments section
  • .comments, .comment_area, .responses e.g.
• Should keyboard shortcuts (left/right) be a separate browser extension?
• Automated tests

Note: webtoons uses .paginate for both the main nav and pagination of the comments area.

• The ability to choose where to save images.
• Add waifu2x Artwork/Scans Model
• Reduce the number of requests (there should be an opinion of 50 requests in 5 minutes) while all photos should be processed at once.