x264 presets in lossless(?) conversion

[Juha]

Will different presets in x264 affect the video quality, if I'm using CRF value 0?

For example, will ultrafast preset produce as good quality as placebo, with CRF 0?

[KarMa]

Should just affect the file size, faster settings giving bigger files.

[smitbret]

In general, your CRF is what determines the quality. 0 should be the same, regardless of what speed you used.

I refer you to Matt Gadient's Guide and his explanation of:
-Encodes fast
-High quality
-Smallest file size possible

Pick 2


https://mattgadient.com/2013/06/12/a-best-settings-guide-for-handbrake-0-9-9/

[_Al_]

set --qp 0 , not --CRF 0

YUV video to get High 4:4:4
x264.exe --qp 0 --keyint 30 --profile high444 --output out.264 input_yuv.avs
-GOP is 30 frames long, you do not want to have it default 250 frames long, not sure what is the reason to have lossless H.264
-to specify profile high 4:4:4 might not be necessary and it is created automatically

[Juha]

set --qp 0 , not --CRF 0

YUV video to get High 4:4:4
x264.exe --qp 0 --keyint 30 --profile high444 --output out.264 input_yuv.avs
-GOP is 30 frames long, you do not want to have it default 250 frames long, not sure what is the reason to have lossless H.264
-to specify profile high 4:4:4 might not be necessary and it is created automatically

I'm using Ripbot264 and I'm not very familiar with different command line options in encoding. Ripbot doesn't even give an option "--gp".

[Juha]

I made 2 encodings with CRF 0, using both ultrafast and placebo presets. The output looks like exactly same for both.

First sample:

Ultrafast:

http://i.imgbox.com/U5jRtBwX.png

Placebo:

http://i.imgbox.com/Y1jSPRpl.png


Second sample:

Ultrafast:

http://i.imgbox.com/Ps6E6Xug.png

Placebo:

http://i.imgbox.com/AfjtYzQl.png


Third sample:

Ultrafast:

http://i.imgbox.com/FrnFywzy.png

Placebo:

http://i.imgbox.com/ERkWyowe.png

[LigH.de]

"Looks like" is no valid criterion for "being lossless". Did you calculate the difference between both pictures, possibly with contrast enhancement, to check whether they are really identical?

If one of both videos would have a worse quality, then the conversion would not be "lossless". If both conversions are "lossless", their quality must both be the same, 100%.

A "constant quantizer" mode with value 0 (--qp 0) will provide lossless compression (down to the color space and chroma subsampling). A "constant rate factor" (CRF) mode will not guarantee this to the last bit.

[Juha]

*double*

[Juha]

Sorry my previous comment, it looks like Ripbot264 allows to customize options after all.

Is this ok? I added "--gp 0" there.

--level 4.0 --gp 0 --preset faster --aud --nal-hrd vbr --vbv-bufsize 25000 --vbv-maxrate 25000

[LigH.de]

q (from "quantization") is a lower case Q, not G.

Furthermore: Forget lossless encoding and VBV. Your bitrate will explode. If you want lossless encoding, expect huge results, certainly not compatible to any consumer video format.

[Juha]

q (from "quantization") is a lower case Q, not G.

Oops, thank you for pointing that out!

Furthermore: Forget lossless encoding and VBV. Your bitrate will explode. If you want lossless encoding, expect huge results, certainly not compatible to any consumer video format.

I don't mind even if my settings will result a huge file. My intention is to edit some video and then upload it to Youtube, and since Youtube will always re-encode uploaded videos, I want to edit losslessly. I have a good upload bandwidth and I'm going to delete the output file after I have sent it to Youtube.

Thanks for help!

[LigH.de]

In this case...

Really lossless quality is not necessary, "visually transparent" will be enough (CRF around 6..12 is good enough not to see a difference, yet get some compression still).

But what's quite important for YouTube is to upload a high resolution. FullHD 1080p should be a minimum to achieve still a convenient playback quality in lower resolutions, reencoded by YouTube. Even upscaling to an UHD resolution may give some quality gain. Look for more existing threads about that topic.

[smitbret]

q (from "quantization") is a lower case Q, not G.

Oops, thank you for pointing that out!

Furthermore: Forget lossless encoding and VBV. Your bitrate will explode. If you want lossless encoding, expect huge results, certainly not compatible to any consumer video format.

I don't mind even if my settings will result a huge file. My intention is to edit some video and then upload it to Youtube, and since Youtube will always re-encode uploaded videos, I want to edit losslessly. I have a good upload bandwidth and I'm going to delete the output file after I have sent it to Youtube.

Thanks for help!

I don't think you understand how much bigger it would be to use the "0" setting. It will be so large that it will be almost unusable, let alone uploadable. If you want true lossless then just remux and forget about re encoding altogether.

[LigH.de]

Indeed; x264 is a quite efficient compressor, it is possible to achieve a convenient result with a compression ratio of about 1:1000 (compared to uncompressed YUV 4:2:0 a.k.a. "YV12"). But lossless video codecs hardly reach a compression ratio of 1:5 ... that means, you would have to deal with at least 200x the size you were used to.

[Juha]

Thank you for your tips.

My source material is DVD quality. If I edit it in a lossy way (let's say I use CRF value 6) but I also upscale it to Full HD resolution, would it be better on Youtube, than losslessly edited video at dvd resolution?

[sneaker]

Yes.

[KarMa]

A "constant rate factor" (CRF) mode will not guarantee this to the last bit.

I was under the impression it did but could be wrong.

Anyway here are the subtraction photos done in avisynth, and each pair of images provided by OP have the exact same CRC-32 hash.

#example of script used on each subtraction image
A=ImageSource("...Ultrafast_2.png")
B=ImageSource("...Placebo_2.png")
Subtract(A,B)

[Brad]

Yes, --crf 0 is lossless. At least with 8-bit x264. I see some old posts saying that 8-bit input to 10-bit output will not be lossless with --crf 0.

[KarMa]

For fun I took the frog image and then converted it to JPEG using different quality settings, then subtracted the JPEG from the PNG source.

JPEG Quality=100 in Paint.NET (4:1 Compression ratio over the PNG)


JPEG Quality=50 in Paint.NET (32:1 Compression ratio over the PNG)


Should point our the PNG image is already losslessly compressed and so skewing down the compression ratio, which would be higher compared to raw BMP.