Believe it or not, someone asked me to make this.
Reencodes of Beatrice-Raws/Kawaiika-Raws with [my full batch subs](https://nyaa.iss.one/view/1582135). Should look significantly better than existing encodes of [similar size](https://slow.pics/c/1r4bUCWj). But obviously you should download [MTBB](https://nyaa.iss.one/view/1582135)—or better yet, [OZR](https://nyaa.iss.one/view/1582362)—if you want ideal video quality.
These are AV1 10-bit encodes, so it may be hard for many users to play them without dropping frames. You should be okay **if** you have a fully updated version of your chosen media player, whether it's [MPC](https://github.com/clsid2/mpc-hc/releases/tag/1.9.21), [mpv](https://mpv.io/), or [VLC](https://www.videolan.org/vlc/).
Encoding settings: `av1an -i owariTwo06.vpy -e svt-av1 -v "--rc 0 -q 22 --lp 1 --keyint 999 --enable-overlays 1 --enable-tf 0 --tune 0 --preset 3 --input-depth 10" -x 250 -w 10 --min-scene-len 12 -l my_log -o owariTwo06.mkv`
6mo-later update: aomenc-lavish tune-ssim preset cpu5 now similarly matches the speed of svt-av1 and gets better ssim2 scores.
svt-av1 cannot match the quality and efficiency of aomenc-lavish preset cpu3
For the newer minis I'm using vanilla aom cpu 3.
I don't know a lot about the lavish fork. Are there AWCY results or something similar that I can look at? My current perception is that it makes a lot of nice changes that help achieve better fidelity, but fidelity isn't really the goal of minis, so...
I’ve seen people post AWCY results in the AV1 community discord and the AWCY discord, never looked into them myself. I can say that using aom-lavish --tune-content=psy with --tune=ssim produces better ssim2 scores than mainline aom. I can also say that being able to use more than 35 lookahead is nice too.
I wouldn’t be surprised if using aom-lavish with --tune-content=psy, --tune=ssim, -cpu-used=5, add 4 to the cq, could produce results that are similar to aom-mainline cpu3, but with the added benefit of being encoded three times faster.
Of course we’re always willing to encode slower for the sake of quality; psy+ssim+cpu5 are appropriate depending on the content.
Update on what I said regarding lookahead: some av1 community discord members are running benchmarks including ssim2 (and AWCY) this week: lag in frames (lookahead) 64 had a very minor effect on ssim2 scores.
I would think that any AV1 encoder who was focused on quality would want to disable TPL anyway, which I imagine would make lag-in-frames past 35 irrelevant.
Wasn't that issue with bad arnr default settings fixed in recent aom vanilla builds?
EDIT: Guess it wasn't fixed completely since I still see a bit of blocking on keyframes in high-motion scenes. I guess it doesn't matter since there's no theoretical reason to enable denoising on reference frames in pre-denoised anime encodes anyway, so I should turn it off.
Comments - 33
Gnome
LightArrowsEXE
halofan
herkz
Bakugo
Oosik
motbob (uploader)
hchcsen
Computron
seth001
neoborn
Interruptor
xikarra
motbob (uploader)
Interruptor
XeroTec
ChadXBeriz
NekoTrix
motbob (uploader)
LastBreeze
motbob (uploader)
LastBreeze
motbob (uploader)
LastBreeze
LastBreeze
motbob (uploader)
LastBreeze
motbob (uploader)
LastBreeze
motbob (uploader)
LastBreeze
JohnTitor228
Uchiah