Thread 107136762

PSA for VP9: single pass encodes do not utilize alternate reference frames.

ffmpeg -h encoder=libvpx-vp9
-auto-alt-ref <int> E..V....... Enable use of alternate reference frames (2-pass only) (from -1 to 6) (default -1)

H264:
PARAMS: -c:v libx264 -preset slow -crf 27
encoder: Lavc61.19.101 libx264
frame= 1253 fps= 99 q=-1.0 Lsize= 3425KiB time=00:00:52.12 bitrate= 538.2kbits/s speed=4.14x

VP9:
PARAMS: -c:v libvpx-vp9 -b:v 0 -crf 38 -cpu-used 1 -tile-columns 1 -row-mt 1 -enable-tpl 1
encoder: Lavc61.19.101 libvpx-vp9
frame= 1253 fps= 27 q=28.0 Lsize= 2658KiB time=00:00:52.20 bitrate= 417.0kbits/s speed=1.11x

2-pass VP9:
PARAMS: -c:v libvpx-vp9 -b:v 0 -crf 40 -auto-alt-ref 1 -cpu-used 1 -tile-columns 1 -row-mt 1 -enable-tpl 1 -pass 2
encoder: Lavc61.19.101 libvpx-vp9
frame= 1253 fps= 12 q=40.0 Lsize= 2336KiB time=00:00:52.20 bitrate= 366.5kbits/s speed=0.502x

AV1:
PARAMS: -c:v libsvtav1 -preset 8 -crf 35
encoder: Lavc61.19.101 libsvtav1
frame= 1253 fps=102 q=36.0 Lsize= 2619KiB time=00:00:52.16 bitrate= 411.3kbits/s speed=4.26x

I was able to find a lossless 720p24fps video source, about a minute long but 1.6GB in file size.
https://media.xiph.org/video/derf/y4m/sintel_trailer_2k_720p24.y4m

Keep in mind that these encodes are between VMAF 90 and 95 which seems to achieve a good ratio of compression and quality. Makes sense to use the H264 one as a baseline to compare the other ones.

Filesize= 3.34 MB
Bitrate= 535 kbps

Okay what is this thread for?
Are these the best settings for best file size with still very good picture quality?

>>107136906
Video codec autism, mostly. I just found out right now that VP9 will not utilize alternate reference frames in single pass mode, which hurts efficiency.

VP9 (1 pass)
Filesize= 2.59 MB (-22%)
Bitrate= 416 kbps (-22%)

VP9 (2 pass)
Filesize= 2.28 MB (-32%)
Bitrate= 365 kbps (-32%)

>>107137004
Okay if you can figure out which one is the best for file size and quality, I would like to know.
I have tried AV1 and I wasn't impressed by the results the intel encoder should be way better than that.

OFC AV1 is still banned here. I don't know why I kept my hopes up.

AV1
Filesize= 2.55 MB (-24%)
Bitrate= 409 kbps (-24%)

>>107137079
VMAF scores between 90 and 95 seem to be the sweetspot and you can easily verify this with ffmetric GUI if you can't use the standalone vmaf executable. If you mean encoder settings then SVT-AV1 is king for that as it encodes faster than x264 while providing smaller filesize compared to H264 on the default preset 8/10 already. If you have time preset 5/6 can net you another 10-20% lower filesize. Throwing 10-bit in there will get you another 10-20% lower filesize but tank your FPS so you probably shouldn't use that unless you have a lot of time on your hands.

>>107136762 (OP)
lol who does a single pass of that terrible encoding?
you can always tell the webms of vp9 single pass. pure shit all huge blocky artifacts.

worthless midwit thread
>>107136762 (OP)
>-auto-alt-ref 1
use 6
>-cpu-used 1
use 0
>-tile-columns
don't do this, lowers quality
>-row-mt 1
fine if actually needed


i rarely drop trvth nvkes but this is the endgame template for the most efficient highest quality two pass vp9 encoding:
ffmpeg8.exe -i .\input_name.mp4 -an -pix_fmt "yuv420p10le" -auto-alt-ref "6" -arnr-maxframes "0" -arnr-strength "0" -lag-in-frames "25" -enable-tpl "1" -c:v "libvpx-vp9" -b:v "[use your FUCKING brain and a calculator]k" -filter:v "scale=[see below]" -g "999999999" -speed "0" [also known as cpu-used] -passlogfile "the_input_name" -pass yadayada

here are roughly appropriate bitrate ranges for corresponding resolutions (scale):
https://developers.google.com/media/vp9/settings/vod/#bitrate

albeit very slow, but i sincerely hope you care about sending timeless quality to the world

read docs


SAGE!!! stupid niggers

>>107137221
I used to but c'mon man that's 90% of VP9 encodes here. Nobody told how how shit google's joke video codec truly is.

>>107137260
#RUDE

10-bit nukes FPS and 6 auto-alt only works for that not 8-bit. Also encoding will fail unless you specify profile 2 because google DID NOT make 10-bit part of profile 1 because... fuck knows man. Also means like 50% of phoneposters won't be able to see your webm here because google did not enforce profile 2 VP9 support. AHHHHHHHHHH.

https://www.webmproject.org/vp9/profiles/

Also
>bitrate
lol you think you can do better than the CRF algorithms?

>>107137260
I disabled tile columns and used cpu-used 0. It barely made a dent, cleetus. Not sure why I'd want to encode at 5 FPS for like a 0.2 improved VMAF score. This is no different than using veryslow instead of slow with x264. I can't see this making a difference even if I encoded TV shows desu.

>>107137387
>>107137535
your reading comprehension is niggerlicious, you're unintelligent, and you lost

>>107137387
>Also encoding will fail unless you specify profile 2
NTA but nope, specifying profile is redundant, ffmpeg/vp9 automatically applies whatever profile is compatible

>>107137634
Anything less than 0.5 VMAF can literally be attributed to a measurement error. If you don't trust VMAF, look at the SSIM

0.9901 vs 0.9904

>>107137669
Google took more than a decade to fix this. Amazing.

>>107136762 (OP)
>single pass encoders
You cannot be this degenerate.

>>107138481
The difference doesn't seem that big desu. Then again the goal is to squeegee as much quality out of the puny 4MB file size limit here. Explains why some webms here look better than anything you find on youtube.

Bump because there's nothing good on /g/ right now.

>>107137682
what are the filesizes for both? i see 365kbps vs 340kbps, so a 7% better filesize at a (slightly) higher quality may make sense if you're ok with slow encoding in favor of max efficiency