[gmenzel]

There are already several threads discussing the two blending modes for material layers and how to use the additive mode to produce glossy reflections. But most of these threads seem to center on questions concerning render speed and noise. My focus here is the question of physical realism, which, after all, Maxwell is all about.

Current best practice to produce glossy materials seems to be to stack a low-roughness additive layer on top of a high-roughness base layer. This way you get bright reflections while also preserving the brightness of the diffuse base layer. But this is not how actual physical materials behave, is it?

As I understand it, the additive mode adds the BSDF solution of one layer to that of another layer, thus increasing the amount of reflected light – potentially above the amount of incident light. This would mean, that in some situations (e.g. self-reflections) the reflection of a light source could appear brighter than the light source itself, which of course is not physically possible.

I’m aware that V3 now limits the total amount of energy for added layers to at least avoid numerical problems that can cause render defects. But despite this, strictly speaking, is it not true that any material that uses additive layers will NOT produce results that can be called physically correct?

The user manual (even V3) seems to confirm this by explaining how additive layers introduce bias and that all characteristics of a homogeneous material should be contained within a single layer. It further seems to imply that layers are really only intended be used to create composite materials where two or more materials are mixed together via masks. Is this information outdated?

I ran into this question while trying to reproduce a glossy white leather material from photographic reference. Is there a physically correct way to produce such a material without resorting to the additive layer technique?

Any thoughts on this are appreciated.

[choo-chee]

I think in real life a leather is made from some thin layers anyway ...

[tom]

It's true the Additive Blending wasn't able to preserve the energy in the past until now. So, now it's pretty safe to add layers over and over within the rules you mentioned. We will check the parts of manual and keep them up to date ASAP. About physical correctness, you shouldn't think stacked layers like physically existing layers. It's just an analytical way of preparing the overall surface behavior and surely physically correct.

[choo-chee]

that is bringing me back to an old request (I'm still on 2.7 but this is still I think the most need upgrade in MR):
make a really good material library, so users will not need to know too much... just pick and render.

[AlexP]

It reminds me this:
http://www.luxrender.net/wiki/LuxRender ... ls_Layered

I have question here: will in real world "gloss coat" darken "diffuse layer" beneath it on incident angles?

[gmenzel]

Thanks, Tom. But I still don’t understand how the behavior of materials with additive layers can be consistent with a truly unbiased approach.

It might very well be that all light reflected from a surface will then be treated in a physically correct way. The issue I see is the fact that too much light gets reflected in the first place. Below a simple demonstration:



The scene consists of a torus surrounded by a sphere. The torus material is a basic 225-white lambertian base layer with a mirror perfect gloss-layer in additive mode on top. The sphere (faces inverted) has an emitter applied to it. There are no other light sources in the scene. I intentionally chose the emitter intensity so that the background appears as gray. The rendered result clearly shows the effect the additive layer has. Note the bright self-reflections on the inside of the torus. With every self-reflection bounce the energy increases and the reflection gets brighter

This should not happen. With a physically correct material, no part of that rendering should be brighter than the surrounding light source. Doesn’t this proof that results will no longer be unbiased (i.e. physically correct) when additive layers are used?

[Mihai]

I guess you used a very high nD and force fresnel off to get that result? If you want to keep specular reflections more physically correct, use force fresnel and limit your nD to something between 1.4-1.6. That's for virtually all kinds of plastics.

Generally speaking, you have to realize that there is no way one material model can mimic the many substances found in nature 100% accurately. It is not realistic to expect that. Imagine a wood material has things going on with its reflections due to its surface structure that would warrant its own complete material model. Imagine bread, how would you make a material that renders that 100% accurately? For our purposes what is important is that we have a model which realistically predicts how light will reflect off a solid (non SSS) surface with an even microfacet roughness, or irregular larger bumps. There are no shortcuts or interpolations going on (read: guessing). Regarding plastics, the more accurate result will be if you use SSS, and thus you don't need this additive layer approach. So for plastics, SSS would be the most physically accurate way to go. But you still have to watch out to specify the "right" nD, and not go wild with very high numbers. The additive approach can still be useful when the SSS effect would be negligible and also of course it renders faster.

So really, try to think you have two major material categories: metals (which can be mimicked very well with one BSDF), plastics (which can be mimicked very well with SSS, or alternatively use base layer + additive layer approach which will do the job in many cases and faster).

For the other "stuff".....you get into the area where the complexity of the surface structure contributes so much to the look, that it would require completely separate, and highly specialized material models to mimic that in a physically accurate way so to speak. But we can get away with using the generalized model again for many of these, partly because we can take a photo of the texture resulting from that complex surface, and at least there is no guessing going on in the light transport model.

[photomg1]

Watched a recent lecture by Dario Lanza where his approach to a plastic was back to the old mixing of two BSDF's in a single layer. (I did scream a little inside )

Just wish there was some consistency of the correct approach from you guys .The docs really need to reflect the current stance and it would seem on this aspect they could do with an update.

Slightly off topic ... Love the crayons render Mihai !!

[eric nixon]

I want to know which real-world surfaces have additive reflection and which dont? I always understood that a laquer coated surface would have additive reflection, but how does that transition into a normalized reflection as the laquer becomes worn and scratched?

A better example to discuss might be plastics or leather, which can have a highly polished clear topcoat (additive) or can be scratched/duller but still quite shiny, and then rougher still.. So how should we think about this transition? Is there a transition? Are reflections only additive at grazing angles? (No harm in asking right? I really dont know..)

Is there a link online somewhere to the actual science about 'reflection contribution' regarding material finishes, I tried google but came up empty.

mixing of two BSDF's in a single layer

That still works great for many plastics, and is still my default option for most materials.

[Mihai]

Honestly I think for our purposes the best we can do is take several photos of a sample under different angles, in controlled lighting conditions (with high CRI lighting) and properly white balanced. Then recreating that scene virtually and trying to build the material. The most difficult I think is getting the Maxwell material to react as close as possible to the sample, in terms of reflections, under different angles. Then you'll have a great looking, predictable material.

As for when additive, when not, I think always additive when trying to mimic a topcoat/laquer on top of something else. Because in the real world, say you have a yellow plastic with and without a topcoat. Would the version with the topcoat have a duller and darker looking yellow? I'd guess no. So the topcoat never really takes away from the saturation of the "texture" in my opinion. Duller vs stronger, that's just a matter of finding the right nD, having a good roughness map etc. That's where good references taken in controlled conditions come in handy.

Slightly off topic ... Love the crayons render Mihai !!

Thank you! I will do a better one where the paper is thicker and has SSS

[photomg1]

Well if you do find any great links please share Eric !
I for one would love a deeper understanding.

Best

Matt

P.s
Look forward to the revised sss render Mihai : )

[gmenzel]

I want to know which real-world surfaces have additive reflection and which dont? I always understood that a laquer coated surface would have additive reflection, but how does that transition into a normalized reflection as the laquer becomes worn and scratched?

Because in the real world, say you have a yellow plastic with and without a topcoat. Would the version with the topcoat have a duller and darker looking yellow? I'd guess no. So the topcoat never really takes away from the saturation of the "texture" in my opinion.

There is no such thing as additive reflections in nature. The amount of reflected light can never exceed the amount of incident light.

Light that is reflect diffusely by one layer can then not also be reflected specularly by another layer in addition to that. That would violate the law of conservation of energy. The rule should be something like this: Total amount of light reflected = Total amount of incident light – Amount of light absorbed (leaving emitters and sss out of it for now).

Only in cases where the base layer is purely black at 100% roughness will having an additive layer not introduce bias. In all other cases you will end up with a material that acts as a light amplifier.

Please don’t get me wrong (this is in regards to what Mihai wrote), I do understand the need for shortcuts and why certain compromises are unavoidable in face of the unfathomable complexity of real-world surfaces. I grew up with biased renderers and am used to thinking of materials in terms of mere imitations of the real thing. In this regard I appreciate having the option of additive layers. My only criticism here would be that the ins and outs of this approach may not be conveyed properly to the user at the moment.

And if we still are at that point where hacks are unavoidable to achieve realistic-looking results, I would wish for some more options within Maxwell’s material system. What I miss most at the moment are falloff maps to blend layers and BSDFs. My initial reason to start this thread were my failed attempts at producing a leather material that incorporates proper shallow-angle reflections (i.e. micro roughness). A falloff map would solve that problem. But resorting to additive layers seems to be the only way to simulate such reflections at the moment. Unfortunately, the self-reflection problem as mentioned above makes such materials rather context dependent.

[Half Life]

I try to stay out of these conversations because I don't want the forum to devolve into needless drama -- however it should be said that for your usage case this blog post would likely be a good start:

http://blog.maxwellrender.com/tips/push ... materials/

The practical issue with using additive layering is it does not exist -- for example: in a real-world plastic material with a clear-coat, the clear-coat is a separate material all its own -- with a real measurable thickness, and its own (completely separate) surface properties. Until we can create a clear coat with a real-world thickness and surface properties, making materials in Maxwell will always be more complex than it could to be. From a programmers view that may be mostly an issue of terminology, but for a software that prides itself on using real world terminology (in so many ways), it is always frustrating when the terminology breaks down and the user ends up dealing with terms and concepts that only exist in CG-land.

[eric nixon]

Hey guys what about the transition between additive and not? Really want some examples about how that should work?

[Mihai]

There is no such thing as additive reflections in nature. The amount of reflected light can never exceed the amount of incident light.

And that shouldn't happen with additives, at least since V3. You should be able to use as many additive layers as you wish now, without your object starting to glow or look overly bright. Your test is puzzling, and checking with an SSS equivalent, I get more what I expect....

Yes I agree angle based falloff maps would add an extra option of flexibility.