"Film" Tune for x265 - Please Help!

[FoodOcean]

Up until recently, I have encoded all of my live-action content using the following settings: x264 8-Bit, CFR20, Preset=Very Slow, Tune=Film. While I know it isn't the most accurate setting, I liked the texture/liveliness it gave to my content, and the quality was high enough than my wife and I couldn't tell the difference at normal viewing distances.

However, as I add more movies/shows to my collection, and with storage prices being as insane as they are right now, I have been strongly considering moving to x265 10Bit, as I feel it is the best balance of compression efficiency, quality, and compatibility. The only problem is that there is no native/built in "film" tune like x264.

This wouldn't be a problem, as x265 can achieve higher quality and more detail at similar or lower bitrates; however, the default tune is too smooth looking to my eyes, and the "grain" tune inflates the file size so large that I might as well not encode at all! I am hoping for find some assistance/guidance form this forum that will help me home in on the right settings to craft the perfect "film" tune comparable.

For context, I am working with 1080p remuxes. Encoding time isn't super important, as I primarily run handbrake overnight, but I would like to be able complete at least 3 encodes of feature length films while I sleep. I want quality to be equal to or greater than my original workflow, but I want my bitrate/file size to be at lease 15-20% small, otherwise what's the point, right? And again, I know there is no "one size fits all" in encoding, but I intend to use this resulting setup for all my live-action content.

I have experimented a lot with SAO, AQ-mode/strength, Psy-rd/rdoq, and qpstep, but I feel like I am missing something. Can you lend a hand?

Bonus Questions: why isn't there a "film" tune for x265; Is it because it was designed for 4k and so the loss of detail/sharpness is as big of a concern at higher resolutions? Also, does anyone know the equivalent of x264's psy-rd value of 1.00:0.15 in x265's Psy-rd/rdoq values? Finally, when using FFmetrics to measure encode score, is it better to have PNSR and SSIM decrease slightly to have VMAF increase a moderate amount, or have all 3 metrics increase, but only slightly?

[GeoffreyA]

As a first step, start with:

* no-sao
* deblock -1 or -3
* aq-mode 1
* ctu 32
* rskip 2 and rskip-edge-threshold 1
* preset slow or slower

If still too soft, raise psy-rd and psy-rdoq to 3 or higher, seeing if it helps. Bitrate will increase.

These are my settings for 1080p content encoded on a comparatively slow 6-core CPU. However, I still had loss of detail compared to x264. So, at present, I am in the process of adjusting the psy-rd and psy-rdoq values to compensate, but haven't got them quite right yet.

x265 --output-depth 10 --level-idc 41 --preset slow --crf %crf% --no-sao --deblock -3 --aq-mode 1 --cbqpoffs -3 --crqpoffs -3 --ctu 32 --tu-intra-depth 3 --tu-inter-depth 3 --limit-tu 4
--rd 4 --psy-rd 3 --psy-rdoq 3 --rskip 2 --rskip-edge-threshold 1 --ref 4 --limit-refs 3 --subme 4 --max-merge 4
--no-open-gop --keyint 240 --min-keyint 24 --rc-lookahead 240 --bframes 8 --no-b-intra --weightb --fades --no-amp --dither
--sar 1 --colormatrix bt709 --transfer bt709 --colorprim bt709 --range limited --chromaloc 0
--y4m --input - --output "%dst%"

[microchip8]

@GeoffreyA

add qg-size=16 (for HDR only)

To increase perception of detail, subme=7 does it. It doesn't sharpen but finds the best mv that leads to the perception as if it's been slightly sharpened. Personal experience here, but my CPU is faster than yours and is overclocked to 4.2 GHz

[GeoffreyA]

Thanks, microchip. The thing is, even I-frames are softer, and they aren't affected by subme, as I understand. So far, psy-rdoq is helping. About 3 or 4 seems to be the threshold where it becomes "more detailed" than the source. But I've tested only single frames, and there are softer patches, owing to AQ.

I'd use subme 7 but it does slow down encoding a bit. Presently, encodes run around 0.45x real time. It's a 4600G, and I missed the chance to get a 5800XT when it was heavily discounted. I'll wait for AM6 to upgrade.

As for no-strong-intra-smoothing, detail is less, so I wouldn't be surprised if it's faulty in x265.

[microchip8]

@GeoffreyA

Might be good to point to your previous thread where we discussed a lot of things. I can't find it now, but it has good information

[GeoffreyA]

Got it:

https://forum.doom9.org/showthread.php?t=186697

[FoodOcean]

* no-sao
* deblock -1 or -3
* aq-mode 1
* ctu 32
* rskip 2 and rskip-edge-threshold 1
* preset slow or slower


I will definitely try this, but can you explain to me why you used AQ-Mode=1 as opposed to =3? I know the x265 "Grain" tune sets AQ-mode=0, while the default is =2, so are you splitting the difference? I always though AQ-mode=3 was for dark scene bias and detail retention.

Also, why set RSkip=2? I thought high values meant that they skip more detail, no?

[rwill]

As for no-strong-intra-smoothing, detail is less, so I wouldn't be surprised if it's faulty in x265.

no-strong-intra-smoothing is really hard to do wrong in an encoder so it is most likely implemented correctly. While I personally noticed different detail instead of less detail, well, the difference should be really small.

In certain cases strong-intra-smoothing gives a coding efficiency gain, but its somewhat small. Thats why its in the standard.

[GeoffreyA]

I will definitely try this, but can you explain to me why you used AQ-Mode=1 as opposed to =3? I know the x265 "Grain" tune sets AQ-mode=0, while the default is =2, so are you splitting the difference? I always though AQ-mode=3 was for dark scene bias and detail retention.

Also, why set RSkip=2? I thought high values meant that they skip more detail, no?

AQ1 is uniform according to a simple algorithm of more energy in macroblock == more quantisation. Some argue that it gives better results for high-quality encoding. AQ2 and up use a calculation based on the energy of the entire frame, but it is unclear what the benefits are. AQ3, indeed, comes with dark bias on top of that. In x264, I do use AQ3 as a matter of course; but x265's version, if I remember correctly, raises the bitrate too much. For anime, AQ3 is a necessity.

Recent discussions on AQ modes:
https://forum.doom9.org/showthread.php?t=186641

For grainy content, rskip 2 is thought to be better than the default, older mode of 1, and speeds up encoding quite a bit. rskip 0 would be the best quality but is slow.

[GeoffreyA]

no-strong-intra-smoothing is really hard to do wrong in an encoder so it is most likely implemented correctly. While I personally noticed different detail instead of less detail, well, the difference should be really small.

In certain cases strong-intra-smoothing gives a coding efficiency gain, but its somewhat small. Thats why its in the standard.

I see. Do you think there is something inherent in the standard and its basic techniques that lead to a softer picture, or is it just x265's implementation? I mean generally, not with reference to strong-intra-smoothing.

[Z2697]

Lossy encoding "takes out" high frequency information. Anything can be blurry, and they will, if the budget is low.

[Z2697]

As a first step, start with:

* no-sao
* deblock -1 or -3
* aq-mode 1
* ctu 32
* rskip 2 and rskip-edge-threshold 1
* preset slow or slower

If still too soft, raise psy-rd and psy-rdoq to 3 or higher, seeing if it helps. Bitrate will increase.

These are my settings for 1080p content encoded on a comparatively slow 6-core CPU. However, I still had loss of detail compared to x264. So, at present, I am in the process of adjusting the psy-rd and psy-rdoq values to compensate, but haven't got them quite right yet.

x265 --output-depth 10 --level-idc 41 --preset slow --crf %crf% --no-sao --deblock -3 --aq-mode 1 --cbqpoffs -3 --crqpoffs -3 --ctu 32 --tu-intra-depth 3 --tu-inter-depth 3 --limit-tu 4
--rd 4 --psy-rd 3 --psy-rdoq 3 --rskip 2 --rskip-edge-threshold 1 --ref 4 --limit-refs 3 --subme 4 --max-merge 4
--no-open-gop --keyint 240 --min-keyint 24 --rc-lookahead 240 --bframes 8 --no-b-intra --weightb --fades --no-amp --dither
--sar 1 --colormatrix bt709 --transfer bt709 --colorprim bt709 --range limited --chromaloc 0
--y4m --input - --output "%dst%"

Too many b frames and no-b-intra is not a good thing, sir.

[GeoffreyA]

Lossy encoding "takes out" high frequency information. Anything can be blurry, and they will, if the budget is low.

Still, x265 tends to blur more than x264 outside transparent bitrates.

Too many b frames and no-b-intra is not a good thing, sir.

In one instance, I found that b-intra was the cause of a streak, so disabled it. How many B-frames would you recommend in conjunction with b-intra?

[FoodOcean]

Ok, so I tried it and it is indeed lower bitrate and has higher FFMetrics scores than my old x264 workflow, so that's a plus! However, when comparing them side by side, the x265 recommendation was looked less grainy/textured, and more artifacty (not a word). do you find that issue at all?

[rwill]

I see. Do you think there is something inherent in the standard and its basic techniques that lead to a softer picture, or is it just x265's implementation? I mean generally, not with reference to strong-intra-smoothing.

In the standard, no. And even if there is a tool that would blur a bit the encoder can turn it off. x265 seems to be unbalanced and its developers don't seem to care much in past couple of years.

[microchip8]

I will definitely try this, but can you explain to me why you used AQ-Mode=1 as opposed to =3? I know the x265 "Grain" tune sets AQ-mode=0, while the default is =2, so are you splitting the difference? I always though AQ-mode=3 was for dark scene bias and detail retention.

Also, why set RSkip=2? I thought high values meant that they skip more detail, no?

In addition to what GeoffreyA said about AQ, the x265 devs have changed the params for AQ2 for the worse compared to the ones in x264. So all other AQ modes that depend on AQ2 are somewhat "broken".

[FoodOcean]

In addition to what GeoffreyA said about AQ, the x265 devs have changed the params for AQ2 for the worse compared to the ones in x264. So all other AQ modes that depend on AQ2 are somewhat "broken".

so that means AQ3 is broken as well, since it is basically AQ2 + dark bias?

[microchip8]

so that means AQ3 is broken as well, since it is basically AQ2 + dark bias?

Yes, it isn't optimal. That's why many here recommend to use AQ1. And as GeoffreyA said, it's not very clear what advantages the other AQ modes have

Personal experience with AQ2 in x265 = it smears patches of the image quite often. Same with AQ3. No such things with AQ1 as it's identical to the one in x264

[FoodOcean]

Yes, it isn't optimal. That's why many here recommend to use AQ1. And as GeoffreyA said, it's not very clear what advantages the other AQ modes have

Personal experience with AQ2 in x265 = it smears patches of the image quite often. Same with AQ3. No such things with AQ1 as it's identical to the one in x264

That is so interesting, all the documentation I have found says the opposite, but I guess that why community knowledge is important! do you know if this is still true even after the handbrake update to 1.11.1?

Also, is there anything else I should know about psy-rd/rdoq as well? the default x265 values are 2.0/1.0 and the grain tune values are 4.0/10/0 respectively, so part of me feels like if I am looking for a happy medium, then I should find settings in that range, no?

[Z2697]

That is so interesting, all the documentation I have found says the opposite, but I guess that why community knowledge is important! do you know if this is still true even after the handbrake update to 1.11.1?

Also, is there anything else I should know about psy-rd/rdoq as well? the default x265 values are 2.0/1.0 and the grain tune values are 4.0/10/0 respectively, so part of me feels like if I am looking for a happy medium, then I should find settings in that range, no?

I don't know what you want to confirm, but as long as they are not using different x265 source code there's no difference.

The default value for psy-rd/rdoq are good enough, if you want a bitrate increase at same CRF as a result of increasing them, better just adjust the actual CRF instead of banging on those settings.

[Z2697]

Still, x265 tends to blur more than x264 outside transparent bitrates.



In one instance, I found that b-intra was the cause of a streak, so disabled it. How many B-frames would you recommend in conjunction with b-intra?

1) I'd say you are pre-biased towards x264.
Or you are comparing different frame types.


2) I didn't necessarily mean them in combination...

x264 doesn't even have a switch to let you disable b-intra (intra blocks in b frames).

b frames are lower in quality (usually) and the "reference B" is half way of that.
x264/5 can only use 1 frame in a consecutive sequence of b frames as reference B.
Some b frames in a long sequence will have "farther" reference frames, the prediction quality can be worse, then there will be more residual to code, and the bits it saves may not even be that many.
So more b frames usually means... less quality. But if you are trying to squeeze out very last bit of, em, bits, maybe it can have a little... bit... of benefit.
There's pbratio, but there's also cutree, the blocks less referenced will have lower quality. non-reference b have no referenced blocks, apparentely.
I know there's b-adapt but... IDK, maybe I'm just paranoid.

Combined they make the situation of bad prediction + forced to use inter mode (when intra costs less) more likely to happen and that wastes bits and lowers quality.

This may sound crazy, but the higher quality (not bitrate bound) you are targeting the less b frames you should use.
Even 0, yes. (for x264/5 only, nvenc for example seemingly enters a different (older) mode when bf=0 and makes worse quality.)

[microchip8]

That is so interesting, all the documentation I have found says the opposite, but I guess that why community knowledge is important! do you know if this is still true even after the handbrake update to 1.11.1?

Also, is there anything else I should know about psy-rd/rdoq as well? the default x265 values are 2.0/1.0 and the grain tune values are 4.0/10/0 respectively, so part of me feels like if I am looking for a happy medium, then I should find settings in that range, no?

Documentation is one thing, practice another. I think the devs changed AQ2 thinking that it will improve the result but didn't thoroughly test their modifications and actually made it worse. Not to sound racist or something, but x265 is mostly coded by Indian coders and they're not known to be the best in their field

I think they misunderstand how AQ2 is supposed to work and making their changes, they believe it actually improves it. But in practice - and I've done a lot of testings with AQ2 - it is overall worse than AQ1

IIRC, it was Dark Shikari (a superb x264 coder from the past, who also came up with Macroblock Tree) that came up with AQ2. In x264, AQ2 works as it's supposed to. In x265, it doesn't.

[FoodOcean]

The default value for psy-rd/rdoq are good enough, if you want a bitrate increase at same CRF as a result of increasing them, better just adjust the actual CRF instead of banging on those settings.

So, my goal is to re-create x264's "film" tune for x265. Are you saying that the default psy-rd=2.0 and psy-rdoq=1.0 is good enough/already scratches that itch? Because I though psy-rd helped determine the threshold of what is considered "noise" vs detail, and psy-rdoq determines how much of that detail to maintain.

[FoodOcean]

But in practice - and I've done a lot of testings with AQ2 - it is overall worse than AQ1

and you are referring to how is visually looks or are you referring to its scores on FFMetrics as well? I know that scores aren't everything, and you should 100% do a blind test to confirm findings, but I find it useful for sanity checks and validation.

[microchip8]

So, my goal is to re-create x264's "film" tune for x265. Are you saying that the default psy-rd=2.0 and psy-rdoq=1.0 is good enough/already scratches that itch? Because I though psy-rd helped determine the threshold of what is considered "noise" vs detail, and psy-rdoq determines how much of that detail to maintain.

psy-rdoq is the equivalent of psy-trellis in x264. I have mine set to 3.5 and psy-rd to 3 and am very pleased.

The more you increase these values, the more distortion it will introduce. The doc says that high values match the source's energy better but I've found that they actually introduce some kind of ghosting and ringing around sharper objects.

Make sure to use rdoq-level 1 as it is the most effective for psy-rd/rdoq. If you increase the former, the latter two become less effective.

[microchip8]

and you are referring to how is visually looks or are you referring to its scores on FFMetrics as well? I know that scores aren't everything, and you should 100% do a blind test to confirm findings, but I find it useful for sanity checks and validation.

I don't use scores. I use my eyes. AQ2 in x265 very visibly smears patches of the image quite often. You can see it even if your eyesight isn't that good.

[Z2697]

So, my goal is to re-create x264's "film" tune for x265. Are you saying that the default psy-rd=2.0 and psy-rdoq=1.0 is good enough/already scratches that itch? Because I though psy-rd helped determine the threshold of what is considered "noise" vs detail, and psy-rdoq determines how much of that detail to maintain.

They are not related in that way.

[GeoffreyA]

1) I'd say you are pre-biased towards x264.
Or you are comparing different frame types.


2) I didn't necessarily mean them in combination...

x264 doesn't even have a switch to let you disable b-intra (intra blocks in b frames).

b frames are lower in quality (usually) and the "reference B" is half way of that.
x264/5 can only use 1 frame in a consecutive sequence of b frames as reference B.
Some b frames in a long sequence will have "farther" reference frames, the prediction quality can be worse, then there will be more residual to code, and the bits it saves may not even be that many.
So more b frames usually means... less quality. But if you are trying to squeeze out very last bit of, em, bits, maybe it can have a little... bit... of benefit.
There's pbratio, but there's also cutree, the blocks less referenced will have lower quality. non-reference b have no referenced blocks, apparentely.
I know there's b-adapt but... IDK, maybe I'm just paranoid.

Combined they make the situation of bad prediction + forced to use inter mode (when intra costs less) more likely to happen and that wastes bits and lowers quality.

This may sound crazy, but the higher quality (not bitrate bound) you are targeting the less b frames you should use.
Even 0, yes. (for x264/5 only, nvenc for example seemingly enters a different (older) mode when bf=0 and makes worse quality.)

Quite right! I am biased towards x264.

Your argument is sound; I agree. By definition, B-frames aren't referenced, so receive higher QPs, and on top of that, mb- or cu-tree pushes up QP again.

It doesn't sound crazy that high-quality encoding should limit the use of B-frames. Indeed, back in the DivX days, it was advised to do so. I distinctly remember that from the DivX PDF, circa 2004. Yet, since x264, we've grown used to a high B-frame count being the standard.

[FoodOcean]

Make sure to use rdoq-level 1 as it is the most effective for psy-rd/rdoq. If you increase the former, the latter two become less effective.

the default is rdoq-level=2, right? I will reduce it to level 1 and see how it pans out.

[FoodOcean]

They are not related in that way.

can you help me understand?

[FoodOcean]

I don't use scores. I use my eyes. AQ2 in x265 very visibly smears patches of the image quite often. You can see it even if your eyesight isn't that good.

I see what you mean. It's funny though cause when I compare AQ1 to AQ3, the former looks more textured, but the latter looks sharper/more defined...

do you have an aq-strength recommendation?

[FoodOcean]

Quite right! I am biased towards x264.

Do you use x264 for your 1080p content?

[microchip8]

I see what you mean. It's funny though cause when I compare AQ1 to AQ3, the former looks more textured, but the latter looks sharper/more defined...

do you have an aq-strength recommendation?

I use a strength of 1.1

I've never been able to get AQ3 do its thing in what it claims - improve dark scenes with low energy. Not even in x264 did I notice a big benefit. It just wastes bitrate for not much improvement, if at all

[FoodOcean]

I use a strength of 1.1

I've never been able to get AQ3 do its thing in what it claims - improve dark scenes with low energy. Not even in x264 did I notice a big benefit. It just wastes bitrate for not much improvement, if at all

Thanks, Ill give it a try, but that raises another question: at what point is it better to just increase the bitrate or CRF vs tweaking settings?

[FoodOcean]

What are people's feelings about selective-SAO=2 vs SAO=0?

[excellentswordfight]

What are people's feelings about selective-SAO=2 vs SAO=0?
For what use case? live-action with sufficient bitrate? Turn off SAO, high-compression and/or animation? selective-sao 1 or 2.

So, my goal is to re-create x264's "film" tune for x265. Are you saying that the default psy-rd=2.0 and psy-rdoq=1.0 is good enough/already scratches that itch? Because I though psy-rd helped determine the threshold of what is considered "noise" vs detail, and psy-rdoq determines how much of that detail to maintain.
I recently did an exhaustive test on psy-rd and psy-rdoq, and for both parameters on heavy grain footage, increasing them procedures encodes that derived more from the source than default values.

--profile main10 --preset slow --aq-mode 1 --no-sao --deblock -1,-1 produces an image is comparable to x264 with tune film in terms of characteristics, is there any specific issues you have found/that you dont like with the output?

[GeoffreyA]

Ok, so I tried it and it is indeed lower bitrate and has higher FFMetrics scores than my old x264 workflow, so that's a plus! However, when comparing them side by side, the x265 recommendation was looked less grainy/textured, and more artifacty (not a word). do you find that issue at all?

I did indeed, and that's why I'm not fully satisfied with my settings and x265.

In the standard, no. And even if there is a tool that would blur a bit the encoder can turn it off. x265 seems to be unbalanced and its developers don't seem to care much in past couple of years.

There seems to be quite a bit left in the tank with x265.

=AQ2 in x265 very visibly smears patches of the image quite often.

The danger is that it might make another part of the picture better, but when comparing, we might not see the smeared portion.

Do you use x264 for your 1080p content?

I used to and had no major problems. Of late, I have been trying to migrate to x265 for the sake of future-proofing the encodes. I'd like to use AV1 but the encoders have temporal instability, where, if stepping through frame by frame, strobing is evident. VVenC is worse. In this regard, x264 and x265 are still the best and safest.

Thanks, Ill give it a try, but that raises another question: at what point is it better to just increase the bitrate or CRF vs tweaking settings?

As you saw, the rabbit hole gets deeper and deeper, and at some point, we are splitting hairs. The best advice would be to start off with basic, solid settings, such as turning SAO off, picking an AQ mode, and using a preset, then adjust CRF according to taste and size. Also, if the file-size difference coming from x264 isn't that much, it merits consideration.

[microchip8]

Thanks, Ill give it a try, but that raises another question: at what point is it better to just increase the bitrate or CRF vs tweaking settings?

At the point where you're not happy with all you've done, then just lower the CRF by 1 and see if you are happy then with the same settings but with a lower CRF.

[GeoffreyA]

microchip,

I tried rdoq-level 1 today, per your recommendation, and it was one of the missing links: quality is improved, the detail retention appearing more consistent.

[microchip8]

microchip,

I tried rdoq-level 1 today, per your recommendation, and it was one of the missing links: quality is improved, the detail retention appearing more consistent.

Great! :)

I got it from the docs, which say level 1 is the most effective for psy-rd/rdoq. The other levels make them less effective.

[x264N00b]

What are people's feelings about selective-SAO=2 vs SAO=0?
You can use the defaults and leave it enabled when you are targeting encodes with higher bitrates. The recommendation to disable it is a leftover from x265 builds < 3.5 or so, where SAO was responsible for blurring the image heavily. Without SAO x265 encodes tend to look like x264 encodes but you sill have the disadvantage in encoder speed.

[rwill]

You can use the defaults and leave it enabled when you are targeting encodes with higher bitrates. The recommendation to disable it is a leftover from x265 builds < 3.5 or so, where SAO was responsible for blurring the image heavily. Without SAO x265 encodes tend to look like x264 encodes but you sill have the disadvantage in encoder speed.

Thats not true. You should not use x265's SAO implementation at all.

[x264N00b]

Thats not true. You should not use x265's SAO implementation at all.
Not true in my opinion, but feel open to discuss it. I did many tests and the encode with SAO disabled looks noticable uglier then the encode with SAO, because disabling SAO emphasized the grain. (like sharpening artefacts) Maybe it has something to do with ringing, dunno :)

Example: https://slow.pics/c/06BZXmD2
Look how ugly the non-SAO Encode looks. (Source has 35mbps avg.)

Settings:
--crf 17 --preset slow --output-depth 10 --profile main10 --level-idc 4.1 --high-tier --ctu 32 --subme 4 --merange 26 --no-open-gop --keyint 240 --min-keyint 24 --rc-lookahead 120 --bframes 6 --vbv-bufsize 30000 --vbv-maxrate 34000 --aq-mode 1 --pbratio 1.25 --deblock -3:-3 --no-sao

[Z2697]

Smeared grain is better?
"Emphasized grain" actually has something to do with psy-rdoq, even the default 1 is a bit too high.

[FoodOcean]

I recently did an exhaustive test on psy-rd and psy-rdoq, and for both parameters on heavy grain footage, increasing them procedures encodes that derived more from the source than default values.

--profile main10 --preset slow --aq-mode 1 --no-sao --deblock -1,-1 produces an image is comparable to x264 with tune film in terms of characteristics, is there any specific issues you have found/that you dont like with the output?

Interesting! I was looking at tweaking psy-rd/rdoq values to mimic the difference between x264's default and film tunes, but you are saying that the default values of 2.0 and 1.0 respectively are good enough? do you also lower rdoq-level to 1 as well?

As for you recommended settings, I haven't tried them all together, but I have separately, I will say that I find that deblock=-1,-1 does nothing but inflate bitrate and lower VMAF. I know x264's film and grain tune uses it, but x265's grain tune doesn't, so I am inclined to leave it set to default.

[FoodOcean]

Without SAO x265 encodes tend to look like x264 encodes but you sill have the disadvantage in encoder speed.

for me, the goal is to get the look/feel of x264's film tune, but with the efficiency of x265, so isn't that a good thing?

[FoodOcean]

Smeared grain is better?
"Emphasized grain" actually has something to do with psy-rdoq, even the default 1 is a bit too high.

can you elaborate?

[Hellboy.]

Not true in my opinion, but feel open to discuss it. I did many tests and the encode with SAO disabled looks noticable uglier then the encode with SAO, because disabling SAO emphasized the grain. (like sharpening artefacts) Maybe it has something to do with ringing, dunno :)

Example: https://slow.pics/c/06BZXmD2
Look how ugly the non-SAO Encode looks. (Source has 35mbps avg.)

Settings:

In that example, the one with sao off looks more closer to the source.

[x264N00b]

In that example, the one with sao off looks more closer to the source.
What? Comon... that one with SAO isn't perfect but with SAO off it really looks like trash. Attached Images: Source / SAO / NO-SAO / x264
Edit: okay at 200% zoom both look bad :D

[microchip8]

I also like the no-sao better