Smaller files with SSIM tune

[YaBoyShredderson]

I was using staxrip to do some testing on 4k HDR files. Using x265, i was trying to work out what settings i should use, with the ssim tune, as i did for x264 in handbrake for 1080p blurays.

Its true for both x264 in handbrake, and x265 in staxrip, that the ssim tune produces produces smaller files than using the film or no tune for x264 and x265 respectively.

I thought that using the psy stuff that ssim tune disables makes the encode more efficient? Shouldnt this produce smaller files?

[Sharc]

I think you can't generalize. The '--tune xx' means specific encoder tweaks for optimizing certain criteria. '--tune ssim' is just one of these aiming at achieving the highest ssim metric. Better metrics do however not necessarily mean better visual experience.
'--tune xxx' and psy settings often increase the file size for a given CFR. So when disabled or overruled (e.g. by --tune ssim) the file size will become smaller. It also depends on the source.

Edit:
See the discussion here
https://superuser.com/questions/564402/explanation-of-x264-tune

[YaBoyShredderson]

I think you can't generalize. The '--tune xx' means specific encoder tweaks for optimizing certain criteria. '--tune ssim' is just one of these aiming at achieving the highest ssim metric. Better metrics do however not necessarily mean better visual experience.
'--tune xxx' and psy settings often increase the file size for a given CFR. So when disabled or overruled (e.g. by --tune ssim) the file size will become smaller. It also depends on the source.

Edit:
See the discussion here
https://superuser.com/questions/564402/explanation-of-x264-tune

If the files are bigger they are also higher quality though? I was just confused as to why they were larger as its supposedly less efficent to use ssim.

[Sharc]

If the files are bigger they are also higher quality though?
Normally yes, but ONLY if all other encoding parameters (settings) are identical. Size alone is not a measure for quality.

[benwaggoner]

Note that --tune ssim is a tuning for a particular objective metric. IIRC, --tune psnr and ssim were added by Dark Shakiri years ago because x264 kept losing to encoders tuned for objective metrics in tests for objective metrics. It was sort of a middle finger at people who were hyperfocused on an objective metric instead of actually looking at the video to determine "quality" and to demonstrate that x264 could win any psnr or ssim based evaluation by providing a tuning for that. And also demonstrating that tuning for the standard objective metrics actually hurt subjective quality.

In general, not using --tune ssim is going to deliver better subjective quality at a given bitrate. So, the reducing in bitrate will generally reduce quality more than it reduces bitrate.

The best way to compare this would be a 2-pass encode, so quality at a fixed ABR can be compared head to head.

[Forteen88]

... And also demonstrating that tuning for the standard objective metrics actually hurt subjective quality. ...I wonder, isn't it only parameters like psy-rd, psy-rdo and aq that affects subjective quality, or are there more parameters that does that?
--tune psnr: disables adaptive quant, psy-rd, and cutree
--tune ssim: enables adaptive quant auto-mode, disables psy-rd

[benwaggoner]

I wonder, isn't it only parameters like psy-rd, psy-rdo and aq that affects subjective quality, or are there more parameters that does that?
Oh, my, there are SO many parameters that impact subjective quality.

It's be easier to make a list of features that don't impact subjective quality!