Hey guys. I'm new here so I hope I created topic at the right place.
So while awesome HW was being upgraded to support FFIXPC, I've spent some time exploring what can be done about backgrounds.
In the end I concluded that any kind of pixel-scaling algorithms perform bad on FFIX artwork so I went to try available neural network approaches.
I wasn't surprised that waifu gave (subjectively) very good results on this type of content.
However waifu with average/none smoothing does not produce relevant results and with full-on smoothing... While results look kinda nice and have anime-ish feel to it, the amount of details lost is simply too big.
So I opted to replace waifu's processing with something of my own that (hopefully) will do a better job. Personally, I am satisfied with the results. You can see some examples below.
I haven't done many images yet as I was waiting for a feature-complete release of HW and what I did was mainly quick-n-dirty tests anyway, but here is a quick comparison of the full-scale lossless images.
Main Street WaifuMain Street Mine/PC versionGuardhouse WaifuGuardhouse Mine/PC version For progress updates and (future) downloads visit
this page.