Watashi ga motenai no wa dō kangaetemo omaera ga warui!
Video source: Aikawa https://nyaa.iss.one/view/1492373
Sub source: Rasetsu https://nyaa.iss.one/view/930583
Audio was encodec using Opus 128Kb/s
The command used for encoding is:
av1an -i filename -c mkvmerge -e aom --chunk-method lsmash --split-method av-scenechange --sc-method standard -x 0 --min-scene-len 120 -v “--bit-depth=10 --cpu-used=4 --end-usage=q --cq-level=20 --threads=32 --enable-cdef=0 --kf-max-dist=240 --lag-in-frames=19" -a=”--c:a libopus -b:a 128k"
If you drop frames in scenes were there are lots of movements and you use mpv you can try this https://www.reddit.com/r/AV1/comments/gjq2mv/drastically_improving_av1_playback/ It should help if you can decode on average at more than 1x speed but drop below that on heavy scene, but if you are always below 1x it won’t help much.
A big thanks to Rasetsu for the fansub and Aikawa for the video raws!
Those aomenc settings are completely retarded.
Please never disable cdef at cpu-used <=5 and always use at least a lag in frames of 48.
You'd be better off using more workers than using threads counts that high.
Anyway, here is a great guide that you should follow :
https://www.reddit.com/r/AV1/comments/t59j32/encoder_tuning_part_4_a_2nd_generation_guide_to/
@NekoTrix In fact I already read that post and all the previous works of BlueSwordM, and I even used his for of aomenc, aom-psy. I disabled cdef because it increases decode speed and it doesn't impact output size and vmaf, at least in my test on anime episodes. It's true that using threads it's worse for efficiency(in fact for my previous releases I always disabled them), but recently I changed cpu and I don't have enough ram to fully utilize it with more workers, so I had to enable threads to utilize it at 100%. I made some test and the penality is only of 1-2MB and 0.1-0.2 points in vmaf(that is at 97% for 75% of frames and 93 for 1%). For the lag in frames options, I too have read that higher is better for efficiency, but it didn't turn out to be true in my tests. I tried with many values, from 120(the max for aom-psy) to 16, and with higher values the file size was bigger: for 120 the files were almost 40MB bigger if I remember correctly, all the while the vmaf didn't change more than 0.2 from 120 to 19. The only jump I saw was from going lower than 19, using 18 the files size grows and the vmaf decreased significantly, so in the end I used 19 because the file size was smaller and the encoder was faster.
I even tried all the options that were talked in that thread but almost all made efficiency worse for the obtained vmaf, and the others were situational at best(meaning that in some cases they were better but worse in others); I only tried them for anime thought, I think that for films and similar results may be different(and yes, I tested both aom-psy and the regular aom). Just to be clear, I based my conclusion on file size and vmaf alone, I did some quick visual testing myself but I didn't have time to check many frames(and the ones that i looked into were in line with the result). I know vmaf can be somewhat wrong in some cases but i don't think this is one of those. If you want to share a different command than mine I would be happy to test it, thanks.
Comments - 4
StazCherryBlood
NekoTrix
OpusMaxE
Quantum1248 (uploader)