Vray sss is it really that bad?

[AyrtonSenna]

Hi guys,

I have been reading this in some places from allot of artist preferring the sss from Arnold. Now, im just starting to learn sss, im really a noob on that so I would like to know what you think? why are they saying that? Alsurface looks great, no?
(please let me know if this kind of topic is inappropriate)

check some of the comments.
Would like to learn more about this.

[^Lele^]

There is a random walk SSS shader in V-Ray, called VRayScatterVolume.
It's however just the scatter part, so the setup of a complete shader is somewhat convoluted.
Vlado's putting the finishing touches to its integration directly into the VRayMaterial, thereby offering the complete package.
We're rendering samples for you to look at as we speak, along with comparisons with the same setups on different engines (Arnold, rMan, RedShift), and for different shader variations (i.e. against fastSSS2, ALShader, and so on.).

Stay tuned for more news within the next few hours and days.

[francomanko]

will this work on gpu?

[^Lele^]

Here's the video Vlado just uploaded.
There's a link to a gdrive file (download the mp4 source, don't use the youtube player) with slightly better compression ratio.

https://www.youtube.com/watch?v=aQZP...adimirKoylazov

I'll be sharing some bigger image of that head later on.

[^Lele^]

will this work on gpu?

at some point
yes

One removed from the horse's mouth, so to speak.

[^Lele^]

Attached, a higher resolution render of that Ten24 scanned head Vlado used in the video.
In order, the beauty, the Dome contribution, the Key Contribution, and the shader setup (the UI may change ever so slightly before release.), with a color chosen for the radius, and the color texture plugged into the SSS color slot.
On top of SSS, i added a standard specular with 0.5 gloss and 1.33 IoR, and a Coat layer with 0.55 gloss, and 1.35 IoR, with a blend of 0.5 (i.e. half-strenght).
As the shader is energy preserving, there's no need to manually adjust the specular layers to ensure it doesn't create fireflies.

Later, or tomorrow, a comparison with Arnold and Renderman (i have a few high-quality renders to go through, still.).

[^Lele^]

Attached, a comparison of the same shader with SSS off, and the same map plugged into the diffuse channel (so the specular layers stay identical.)
Further, an AO of the geometry, so you'll see the geometric data the shader works on.

[Muhammed_Hamed]

AyrtonSenna
Take what you see online with a grain of salt, anyone can say whatever they wish specially on social media..
And I see my name in the second part of your screenshot, but at least you tried to hide it (I tried to answer someone's question at this post and there is context to what was said)
I don't like this being shared here directly, there is no point asking if what I said is correct or not on the official forum.
Arnold is included with 3Ds Max and Maya, same as the model I used in that test. Best is to test on your own and come to a conclusion.

About SSS in Vray, Vlado's video answers pretty much every single point I mentioned.. easier controls/workflow and very good results
It is a big milestone in Vray's workflow..

[Xntric]

Im super excited about the upcoming SSS in Vraymtl!

Personally i have been hoping for better implementation of SSS for quite some time now, and i think the criticism given in the facebook comment above was quite fair.

Currently we have

Translucency in vraymtl: singlebounce, confusing and way too many parameters that needs adjustment.
FastSSS: Singlebounce and it seems it has no glossy fresnel
Alsurface shader: seems a bit too intricate for most simple SSS stuff, and i wonder why the 3 layers of specularity wasn't implemented from the original ALsurface.
Volumescatter: Superior quality, but it's quite cumbersome to make blendmtl everytime. It's also not supported on GPU

in my opinion volumescatter implemented in Vraymtl with more simple UI is a BIG upgrade and one that i'm looking forward to a lot.

So good jo, cheers!

[^Lele^]

Here's a link to gDrive to the 28Mb contact sheet that compares Arnold RW and V-Ray's integrated ScatterVolume.
It's three rows, and four colums.
They are all labelled in the top-left corner of each image (small text. zoom.), but the scheme is like so:
V-Ray beauty, SSS, Specular, Colorised Noise Level
Arnold Beauty, SSS, Specular, Colorised Noise Level
The last row contains the absolute difference (or mod(A-B), or |A-B|. I.e. always positive, so always visible.) of the two rows above.

The speed is quite similar for a noise level which is essentially identical ( read the lines referring to max and average. the numbers are close as the render times. We have a slight advantage in noise levels, but it'd need to be confirmed.).
There are a couple of notable differences in the way the SSS solution behaves, between us and Arnold.
a) The scene setup is identical, and the brighter shading seen in the Arnold renders is due to the way its SSS behaves: it will converge to a diffuse BRDF, which isn't quite right for skin. (you can quite clearly see this in the corresponding REs and difference thereof.)
We instead converge to a BRDF with a flatter appearance, not as diffuse-like.
b) With RW there are known issues at grazing angles under some lighting conditions: Arnold seems to converge those areas to white (ish), while everyone else (including us) goes towards a grey (ish).

[^Lele^]

Here's the contact sheet with renderman. (Not the same setup as with Arnold, if it matters to anyone, i'll try and make thre three identical.)
Again, their RW seems to converge to a diffuse-like BRDF.
Other than that, it seems to be darkening more at the edges, for example around the nose (but i may have been setting up the specific mode wrong. The Burley mode was a lot closer to ours, however.).

Two issues here:
a) i couldn't quite match the camera zoom/aperture exactly (I have a special build which works under Max for V-Ray, but have to convert the scene to Maya for rMan.).
b) the noise readouts for V (last number of four, on each line) seem whacky. The color space transform node i use to get luma from RGB seems to have issues in the version i am using of nuke. Disregard those, just gauge averages from the R,G and B values.

Both the above mean the speed is difficult to asses.
Renderman has a slightly lower noise level and slightly higher screen coverage, but it also takes more than twice the time to complete the render.
We seem, all taken into account, a bit quicker, but again it would need (much) better confirmation.

[Muhammed_Hamed]

Thanks for the links and all info Lele ^^
Quite an interesting read for the upcoming weekend, and looking forward to using this new SSS in Vray

Best,
Muhammed

[Xntric]

Thanks Lele. They are quite similar indeed!

I notice the specular pass is a bit cleaner for arnold, but it doesnt seem to be noticeable in the beautypass.
It's gonna be great to have this in vraymtl

[super gnu]

on a slightly related note.. this may tie back to my thread a while ago about rendering a planetary atmosphere. the absolute fastest method i found which included an approximation of rayleigh scattering was the vrayscattervolume. however it proved practically impossible to get a nice feathered edge fading off into space. blendmtl with scattervolume didnt work, mapping scatter radius with falloff/distancetex didnt work/ was impossible to control.. in the end i used a hacky faked approach with gradient ramps etc.

will this new "integrated" scattervolume allow something approaching a (visually) realistic atmosphere to be created? -i.e. can i get a scatter effect i like then simply map opacity in material or something to feather the edge?

[^Lele^]

Yeah, i wouldn't be obsessed about the performance metrics: i am sure i made mistakes setting the scenes up, and a pro of any of the other engines would wipe the floor with me.
The takeaway is indeed that we are very comparable for speed and features, and that we are finally a lot easier to set up and use.

For an atmosphere, i personally think we'd need a proper solution, with Rayleigh and MIE, but i haven't actually tried setting one up with the integrated scatter volume.
Perhaps because i failed with it when it was alone.

Have you considered trying one of the dedicated solutions, like terragen, for the specific project?
They just improved their atmospheric model to include the "Rayleigh phase function" for added realism.