x264 & Youtube upload compression

Can you beat out the quality of the other creators by using a slower x264 preset?

x264 & Youtube upload compression

Have you ever wondered if the different x264 presets would change the quality of the video that youtube shows after their processing?

I certainly have, so I figured I would do a couple of small tests. Sample sizes of ONE whole video.

I was curious about both 4k and 1080p, so I did both.

Found some pretty and free 4k footage, used that and I also encoded a downscaled copy. I encoded it using x264, CRF 16, thru a variety of presets. I chose the preset "veryslow" as my source/flawless copy for the downcaled version. In hindsight this was not the greatest idea, but I still think we can draw some information from this.

Using VMAF to compare our original copy to the other encodes, and doing the same CRF (16), but at different presets, I wanted to see what VMAF would say about the quality of our other encodes. This is the first result, comparing the original 4k footage with our new 4k crf16 encoded copies:

preset      VMAF

veryslow    98.86
medium      98.78
veryfast    97.17
4k encoded at crf 16 comparing our original, very high quality copy. 

Its barely any difference outside of what we would consider margin of error.

Now, for the more interesting part. Lets put these copies 4k up on youtube, and then download the again, and compare them to our "flawless" copy. Does the differences get exacerbated?

Preset		VMAF

veryslow	88.48
medium		88.52
veryfast	87.45
Uploaded the encoded copies to youtube, and downloaded them back down (VP9 from youtube)

No, they do not. They are a bit worse off than what we uploaded, but its the same frankly very minor differences.

Not very interesting, so lets move on to the 1080p encode. These are all compared to the CRF16 "veryslow" downscale. The VMAF differences in the local encodes are as follows:

Preset		VMAF

medium		96.44
veryfast	94.74
More of the same. "Veryfast" falls a bit further back on 1080p

Time to dump them onto youtube and see if there are any interesting bits of information to get there. PS: I forgot to test the "slow" preset, but imagine that its pretty much the same as medium (same as before).

Preset		VMAF

slow		49.31
medium		49.18
veryfast	48.87
1080p vp9 from youtube, comparing to the original pretty downscale

Except for the vp9 bitrate that youtube gave us for this transcode being laughably low, the same low spread exists, not really any change. If anything the spread is smaller, cus it all sucks...

Maybe our good friend h264 (avc) will be more kind:

Preset		VMAF

slow		75.35
medium		74.97
veryfast	73.82
1080p h264 (avc) from youtube, comparing to the original pretty downscale

Actually, yes, in this case the bitrate given to us by our glorious savior youtube did seem to beat out the vp9 encode, but it says very little about the spread. Its a tiny bit more noticeable, but certainly not something I would consider spending the extra time and power waiting around for.

So, in conclusion, is worth spending the extra time having x264 creating a prettier encode at the same CRF value? In my eyes that is a resounding hell no.

2 VMAF "points" is not something anyone would be able to tell apart in a blind test. There is a small difference, so if you absolutely NEEDED to have the best quality possible, there are teeny tiny gains there for you, if you can be bothered to do so.

There is a also a small difference in file sizes to consider, but that has already been done to death, and it just a google search away ;)