Why is CRF encode coming out with banding?

[killerteengohan]

I am using CRF 16 which is fairly good. For some reason, the CRF 16 encode is getting banding or making it worse, while 2-pass looks better.
The settings are the same, except in 2-pass settings, I'm specifying a bitrate, while with CRF, it's choosing it.



The computer monitor in the comparison screenshots is where it's most obvious.

Source vs CRF 16
https://slow.pics/c/ktcMLHyV

Source vs 2-Pass
https://slow.pics/c/9NeeY42u

CRF16 vs 2-Pass
https://slow.pics/c/dNmJKjuS


As you can see, the banding in CRF looks pretty obvious. The 2-pass looks like its getting some too, but its to a much lesser extent and you have to actually look for it.

Whats up with CRF getting this banding so badly? Is CRF 16 not enough bitrate?

I set 2-Pass to 13,738kbps, while CRF 16 seems to be giving it about 8457kbps.

[poisondeathray]

I set 2-Pass to 13,738kbps, while CRF 16 seems to be giving it about 8457kbps.

Compare at the same bitrate and you should get similar results

To put it another way, you're comparing a ~1.6x more bitrate version

[killerteengohan]

So the problem is its not enough bitrate at CRF 16?

But they say CRF 17 should be nearly the same as the source. Thats CRF 16 which is even better, and that is clearly not almost the same as the source.

I've always used 2-pass so I am not an expert on CRF encodes. I just wanted to try CRF to lessen the encoding time.

[poisondeathray]

Yes, the problem is not enough bitrate

When you lower psy-rd, or increase deblock, the bitrate at a given CRF value will be lower

If you like the 2pass results, you would have to use a lower CRF with those settings. It's impossible to "guess" what the correct CRF value should be.

[jagabo]

So the problem is its not enough bitrate at CRF 16?

Yes.

But they say CRF 17 should be nearly the same as the source.

For most people, at normal viewing speed/distance, for most material yes. But there are specific cases where it's not -- like shallow gradients. I think you need to get down to about 10 or 12 to cover edge cases like this. --tune grain might help. Or --aq-strength=1.5 or more. But in the end, they end up using more bitrate.

[poisondeathray]

10bit encoding helps too

[jagabo]

Yes 10 bit encoding can help. How much can depend on your playback device. Some 8 bit devices decompress to 10 bit YUV then convert to 8 bit RGB. Others decompress to 10 bit YUV, convert to 8 bit YUV (simple truncation, no dithering), then convert to RGB. In the former case you get smoother gradients. In the latter case it doesn't help much.

[badyu17]

Source also has banding although less pronounced. This kind of content is perfect for x265 10bit. I mean, those x264 bitrates you use are enormous for this type of content.

Btw, decent show The Spectacular Spider-Man