Nyaa archive 1885360
[LBA]TOHAI.Ura.Rate.Mahjong.Touhai.Roku.01.x264.AAC.720p.mkv
Saved on 2025-01-21 19:26:17. This copy includes the description and comments available at capture.
Description
[LBA] = Low-Bandwidth Anime
Re-encoded video and audio from [Yofukashi] to make file size smaller for those with low-bandwidth internet
So you should only dl this if you live in a region with slow/data-limited internet, or have low storage space
Otherwise, you should just be dl'ing the [Yofukashi] one
Only doing this because there aren't any 720p/small file size alternatives that I know of
Comments 4
Nice~
720p and 71MB is waaaay too low, this is like 5th world country standard.
better 1080p and 1/2 or 1/3 the size of yofukashi
it’s about efficiency and optimization
the anime is a slideshow anyway. doesn't need to be 500MB
This release is ultra bitstarved as though it's for those old .3gp phones. Rather than maxing efficiency in terms of .7zip is better than .zip without losing bits, this release uses the highest possible compression just below the threshold where it would look like VLC crashed.
From my testing you can shave off at least 50% of the filesize with no discernible quality loss to 10bit HEVC, although hitting the sweet spot of settings is difficult. You need a somewhat high psy-rd and psy-rdoq (fine detail) value to retain the film grain in the dark opening scene, which ups the filesize, while setting it too high adds distracting jitter to all the other bright static scenes. Higher crf values (compression) within the acceptable range seem to be fine as the art is simple and high contrast. Maybe the first scene can be encoded separately and joined to the remainder to hit a maximum efficiency with no loss.
ffmpeg -i input.mkv -pix_fmt yuv420p10le -c:v libx265 -crf 21 -preset slow -x265-params "limit-sao:bframes=8:psy-rd=1.5:psy-rdoq=1.5:aq-mode=3" -c:a copy output.mkv is close to optimal, I think, at ~600MB with no/low quality loss, but this is an art I'm not too experienced in.
Gives me more appreciation for MTBB and Deerstalker and the like who presumably re-encode over and over all day to try to hit the sweet spot across scenes in various different lightings, to lose minimum or no detail while cutting away huge amounts of unnecessary filesize, like a challenge.