>>99464 (OP)Here's how it's done. Here's a proper Sharty encode.
This webm is based on this release:
https://nyaa.land/view/1752129The audio is the original mix (sourced from the Laserdisc), included in that release.
Briefly without going into the details of every setting, not even mentioning most settings, basically what was done was this:
Via ffmpeg converted source mkv to an FFV1 (lossless codec) avi while decimating to 23.976 and cropping the 1448x1080 res to 1440x1080, dropped that avi into VirtualDub2 and applied various filters to make the video more compressible and look nicer (as in the quality of the webm - not look nicer than the source obviously; source mkv looks much better) and downsized to 720x540 then saved to an RGB FFV1 avi (RGB as filters were RGB, and ffmpeg's RGB-to-YUV420 conversion is higher quality than VDub2's thus better to let that conversion occur at final encode stage), and then via ffmpeg merged that 720x540 video with the subs from the source mkv, encoded that to another RGB FFV1 avi, then encoded that avi to a VP9 webm (it's a 2pass crf-based bv:0 encode, with row-mt at 0 and every other multithreading-related option disabled, every other quality-degrading thing disabled and every quality-boosting thing enabled, bt709 set via -vf "setparams=range=tv:colorspace:bt709:color_primaries=bt709:color_trc=bt709", tune ssim, yuv420p12le – this is 12-bit, so the vast majority of phonefags can't watch this), then converted the Laserdisc audio from source mkv to a wav and dropped that wav into Audacity and amplified it to -2db peak amplitude, saved it, then via ffmpeg converted that amplified audio to Opus (VBR) as part of a commandline that simultaneously copied the video stream from the silent webm resulting in a new webm with the video and audio muxed together which is what we have here.
Went with 720x540 because the benefits of that res outweighed the quality gained at lower resolutions; did several encodes at 640x480 and 656x492 and 672x504 and overall 720x540 looked better despite needing a higher crf number – the higher res, while resulting in more artifacts in some places, allowed certain details to be preserved better and 540 is exactly half of 1080 so when fullscreening the video on a 1080p monitor it scales exactly 2x.
Of course if the opening and ending & preview had been excluded, quality could've been higher – but the goal here wasn't only to make a better-looking video than OP's (his lacks the ending so could've excluded it) but also to demonstrate it's possible to have it all, the entire episode, at decent quality. The OP and ED & preview is however encoded at a lower quality in order to increase the quality of the rest of the video.
By the way the uploader on nyaa seems like he doesn't know what he's talking about with regards to the fps. He says he IVTCd (meaning inverse-telecining a telecined 29.970 video back to 23.976) which he obviously did not do; the files are still telecined 29.970 (can be seen in VirtualDub2 where the duplicate frames are reported as "[+]" frames), and his VFR talk is nonsensical; Cowboy Bebop is 23.976 (like 99.9999% of anime) and that BD (like the DVDs) is soft-telecined 29.970 and should be IVTCd/decimated down to 23.976 to get rid of the duplicate frames. He wrongly thinks MediaInfo is lying. No, MediaInfo correctly reports the fake/telecined 29.970 fps as "Frame rate: 23.976 (23976/1000) FPS" – MediaInfo knows what it's dealing with, the uploader strangely does not.
