hmm, so the 720p version ended up being larger than 1080p version eh?

Hello, the quality is better than https://nyaa.iss.one/view/957200 ????

the magic of --chroma-qp-offset -4

motbob, what do you mean by that, the increase in filesize is due to the number of chroma pixels is exmendic encode

After compression, chroma subsampling only decreases file-size by 10-20%. yuv444p should only be about that much larger. Not subsampling cannot make a 720p encode larger than a 1080p one at the same settings. RH 1080p Vid: 686 MiB (4.193 kb/s) x264 (10-bit, crf=unk, qp=unk, yuv420p) Aud: AC3 @ 640 kbps (105 MB) RH 720p Vid: 851 MiB (5.211 kb/s; Scum's Raws) x264 (10-bit, crf=15, qp=0.7, yuv444p) Aud: Opus @ 128 kbps (21.2 MB) Resolutions have more of an effect on quality than encode settings usually. Even at these settings, the 1080p one will be higher quality. That excludes filtering of course. Feel free to do a screenshot comparison to check, but that 720p one is bloated. ephemera released a 720p AVC crf=18 one that is 311 MB, or less than half the size.

What I mean is, x264 automatically decreases the quality target of the chroma when it uses 444. exmendic likes to jack the quality target back up, which accounts for some of the increased filesize here. It's beyond my expertise to say whether that makes the files bloated.

It probably increases the file size up to 100MB in my experience. This is one of my old Avisynth projects which are generally a bit bloated. Probably the denoiser (or another filter) did a bad job here, at least I don’t have any bloat issues with Vapoursynth anymore. On my old tests the chroma quality actually benefits from that. However, for future shows I tend to use -2 or -3 instead of -4.

>Resolutions have more of an effect on quality than encode settings usually. Even at these settings, the 1080p one will be higher quality. That’s 100% incorrect. The show’s native resolution is 720p, so with an inverse resizer (like Debicubic, what I used) the settings matter much more. So considering that, the quality here is much higher that with the 1080p version. The studio upscale is Bicubic, which is blurry and worse than any common upscaler you can set in your video player (Spline, Jinc, Nnedi3). Only the ending is produced higher than 720p, just watch the 1080p NCED for that. >ephemera released a 720p AVC crf=18 one that is 311 MB You can’t compare TV encodes with BD encodes. Ofc you don’t encode an Amazon source with crf=15.

@YukinoAi Just in case you don't trust me (how it looks while playback): https://diff.pics/IWvRjcYviFf0/1 (nnedi3 is causing the little shift in MPC-HC, nothing that can be fixed while playback - doesn't really matter anyway) Is the fact that the 720p looks sharper enough reason that "Resolutions have more of an effect on quality than encode settings usually" isn't correct?

eXmendiC, I was generalizing about encode settings which should have been conveyed by using the words "usually [...] at these encode settings." Naturally, I was also not talking about filters. encode settings != filter settings. I never use the two interchangeably. __All else equal__, encoding from the same source, a 1080p encode one will be better quality than a 720p one. Hell, even a losslessly encoded 720p is lower quality than 1080p transcode, all else equal, because the 720p resolution throws away so many pixels. Faithfully representing the few remaining ones (720p) offers a huge quality hit over lossy encodes at the native resolution (1080p) given sane encoder settings. That is what I meant. Is that "100% incorrect"? Now all of that was meant under "all else equal". After agreeing or disagreeing on the point above, then, and only then, can you argue that all else is not equal so the obvious conclusion, 1080p>720p, should not be applied.

As said, there some differences to consider: the 1080p version is not native, unfiltered (overly soft), so you get the whole psycho-visual vs objective "quality" and what "quality really means" conflicts, and that specific 720p encode is filtered. eXmendiC, if a source has a defect, like oversoftening, the solution is not to downscale it, run some filters, encode it using deliberately bloated settings (messing with the chroma quantizer), upscale to compare to the 1080p and then claim it is better quality. Seriously, just filter the original 1080p, and then encode. Done. , that is both subjectively and objectively better quality than an equivalent 720p stream Then you would not be looking at a diff pics of a 720p->1080p with upscaling artifacts that compares a filtered vs unfiltered source where "quality" is a tossup depending on various person-specific factors (subjective quality vs objective, filze-size, psycho-visual).

No, a 1080p won't be better than a 720p with same filtering or encoding settings (except crf, in my experience lowing the value -1 than the 1080p is the way to go) when the shows native resolution is 720p and you inverse downscale it. Is my encode better when I upscale to 2160p? Hell no! And yes, the solution for very soft shows is to downscale it right if they used a soft upscaler like Bilinear. However, if you still believe in 1080p is always better, you are beyond saving.

1080p will always be better than a 720p with same filtering or encoding settings, all else equal. If you still believe in 720p is always better, you are beyond saving.

Okay. Gonna encode everything in 2160p now. Only for the best quality! :^)

Make sure the original source you're upscaling is 480p too for maximum quality. (*•̀ᴗ•́*)و ̑̑

I know I'm late to the party but I made another screenshot comparison between the 1080p and 720p versions during playback for good measure. https://diff.pics/YC-QOKvPfSSv/1 Looks like 720p is definitely the way to go here.