Can the "smoothness" or "blurred" look of x265 encodes be controlled?

[MishaPrada]

I'm a novice at video encoding. Yet, I've found a great x265 downsizing encoding routine for my large television captures.
I can reduce the file size by 40% or more.
But I notice the resulting x265 encode looks noticeably smooth or blurred - with perhaps accentuated edges.
In my current situation, I'm actually satisfied with the result. As I regard it almost like a beautification "touch-up".

But I can imagine potential scenarios where this "smoothing" or "blurring" might not be desirable.
So in those cases, I was wondering if there were any x265 options or commands that could control this aspect of encoding?

[fornit]

There's an x265 option named "--tune" that can be set to "grain". Please find the details at:
https://x265.readthedocs.io/en/default/presets.html#tuning

[MishaPrada]

There's an x265 option named "--tune" that can be set to "grain". Please find the details at:
https://x265.readthedocs.io/en/default/presets.html#tuning

Interesting. So far, I've been encoding live-events and reality-type material.
I don't particularly notice any grain in the source, so I'd probably refrain from activating it there.

This option should be useful for fiction content, where I definitely see grain.

[poisondeathray]

One of the settings that has a large effect on "blurring" for x265 is SAO (it's set to zero in the grain preset) . Many people disable it with --no-sao. But it can be useful for low bitrate, or very clean encoding, or simple cartoons

You usually don't want to use such high --psy-rd and --psy-rdoq settings if not wanting to preserve grain . Much lower than the values used in the --tune grain preset (4 and 10 respectively)

[MishaPrada]

One of the settings that has a large effect on "blurring" for x265 is SAO (it's set to zero in the grain preset) . Many people disable it with --no-sao. But it can be useful for low bitrate, or very clean encoding, or simple cartoons

So I'm encoding high bitrate captures of "reality tv" programming.
One thing I notice is that my x265 compression routine "erases" fine details on skin (like moles & blemishes)...granted, it's mostly in zoomed out shots.
That sort of tells me the "blurring" is a little extreme in my current x265 encoding formula.

You recommended playing with "--no-sao".
And also trying values lower than the grain defaults for "--psy-rd" and "--psy-rdoq"
I'll play around with these values tonight.

If you have any other ideas, let me know.

[jagabo]

Or use a higher bitrate or lower CRF.

[Peter Sagan]

Tune Grain is the best for avoiding blurring and preserving fine detail --given that you allow it enough bitrate to work with--, it is not just for grain, it works with any kind of detail in your source.

[MishaPrada]

Or use a higher bitrate or lower CRF.

I'm not entirely sold on the idea that a lower CRF can fix the problem...because I originally used "CRF 8" for my x265 conversions.
Even in that, I would notice the "smoothness/blurring" effect. Though it wasn't as pronounced.
And all that effort, for a negligible 1-3 GB reduction - on an 8 GB video.

I settled on "CRF 14", because the "smoothness/blurring" produced an interesting aesthetic effect on the type of material I was converting.
But more importantly, I got massive 40%+ size reductions.

Ideally, I'd like to preserve the reduction-levels around that band...but see if I can "cheat the system" by tweaking or filtering to preserve more details.

Tune Grain is the best for avoiding blurring and preserving fine detail --given that you allow it enough bitrate to work with--, it is not just for grain, it works with any kind of detail in your source.

Thanks for the clarification. I should probably start with this simple modification, then fine tune it.

[poisondeathray]

Or use a higher bitrate or lower CRF.

I'm not entirely sold on the idea that a lower CRF can fix the problem...because I originally used "CRF 8" for my x265 conversions.
Even in that, I would notice the "smoothness/blurring" effect. Though it wasn't as pronounced.
And all that effort, for a negligible 1-3 GB reduction - on an 8 GB video.

I settled on "CRF 14", because the "smoothness/blurring" produced an interesting aesthetic effect on the type of material I was converting.
But more importantly, I got massive 40%+ size reductions.

Ideally, I'd like to preserve the reduction-levels around that band...but see if I can "cheat the system" by tweaking or filtering to preserve more details.

Tune Grain is the best for avoiding blurring and preserving fine detail --given that you allow it enough bitrate to work with--, it is not just for grain, it works with any kind of detail in your source.

Thanks for the clarification. I should probably start with this simple modification, then fine tune it.

CRF8 using x265 will usually be larger than the original at default settings (assuming the "original" was some typical compressed source) . There should be no blurring at CRF8 . To put things into perspective, At CRF4, that is mathematically lossless for x265 (it was CRF0 for x264) - at that point it will usually be many times larger than the original . If you got a 1-3GB reduction on a 8GB video using CRF8 - something seems very off

So I'm wondering if something else is going on , or maybe some of your commands are not passing through correctly

How are you capturing ? What hardware and recording format ?

Are you preprocessing your captures correctly ? Are they interlaced ,or telecined, or native progressive ?

Are you using x265 directly or through some GUI, or libx265 through ffmpeg ?

Tune grain will definitely steer you towards retaining fine details, skin pores, fine textures, etc... - that sort of thing . But at a given CRF , it will make the filesize much larger. If you don't have a grainy source, and you don't want to preserve film grain, I would definitely lower the "--psy-rd" and "--psy-rdoq" settings from the preset . Those settings in the preset are only meant to be a "starting point" anyways. It also tends to add edge artifacts when you use higher psy values. Grain usually "covers" that up. If you have a grainless,clean source, you will notice the artifacts more. The compression ratio is also lower, because of ip/pb ratios (the idea there was to minimize grain fluctuations between i/b/p frame quality, and again, it's not as useful for a clean source)

[MishaPrada]

Happy to report, some simple changes to "--psy-rd". "--psy-rdoq", and "--rdoq-level" produced the effect I wanted.
I can confirm, the small mole on Alison Brie's neck is preserved - after x265 conversion.

Source (1080p Television capture, 3 minutes long) = "Bit rate: 43.1 Mb/s, File size: 916 MB"
Past CRF 14 conversion (soft & aesthetically blurry) = "Bit rate: 6827 kb/s, File size: 151 MB"
New CRF 14 conversion (sharper & more textured) = "Bit rate: 10.3 Mb/s, File size: 225 MB"

Thanks for the help guys.
I now have one "softening" routine for sub-par captures.
And one "sharpening" routine for clean captures.

[jagabo]

Happy to report, some simple changes to "--psy-rd". "--psy-rdoq", and "--rdoq-level" produced the effect I wanted.

Why don't you list the settings you used -- they may help someone else in the future.

[MishaPrada]

Happy to report, some simple changes to "--psy-rd". "--psy-rdoq", and "--rdoq-level" produced the effect I wanted.

Why don't you list the settings you used -- they may help someone else in the future.

For the softness routine, it's basically just "--crf 14.00" & desired speed preset.
For the sharpness routine, it's also "--crf 14.00" & desired speed preset...with addition of "--rdoq-level 1", "--psy-rdoq 3.00-6.00", and "--psy-rd 2.0".

To my mind, "--psy-rdoq" is basically the key to the whole thing. I just look at my source & determine how much I can get away with. Then pick something within that 3-6 band.