[AlexP]

gmenzel, can you check your amplification test with roughness 99 not 100 on bottom layer?

[gmenzel]

gmenzel, can you check your amplification test with roughness 99 not 100 on bottom layer?

Attributes of the bottom layer don't influence the amplification behavior. Well, you could set the bottom layer to a pure black. This would of course eliminate the effect. But in any other case too much light will be reflected.

Here a less abstract rendering to hopefully illustrate the problem more clearly. I used a simplified scene with more organic shapes. Both images where rendered to SL16.

The first image uses the additive approach. Just a 225-white base layer with roughness=97% and an additive layer with Nd=1.6, roughness=0 and force-fresnel (same technique as in the video above). Note the bright halos along the contours (as well as the excess noise).


The second image uses just the white base layer with a coating (thickness 0.1mm, Nd=1.6, roughness=0, force-fresnel) applied. This is what it should look like.


Open both images in different browser tabs to see the difference clearly by switching back and forth.

Unfortunately, Maxwell's coatings don't provide a roughness parameter. Therefore this comparison can only be made for perfectly specular reflections.

[Half Life]

I too would find coatings that support roughness a more realistic solution (I always use coatings for anything that is going to be smooth, like car paint, for just this reason) -- not to mention one that makes infinitely more sense from a UI that correlates better to "real-world" recreation mindset.

[gmenzel]

I'm a bit baffled that this problem doesn't seem to get the attention it - at least in my mind - deserves. Not being able to produce materials with rough coatings is a major gap in Maxwells capabilities. Anything that is not metallic or doesn't have high roughness takes trickery to achieve, i.e. to get to look at least plausible.

Sure, you can use a noise bump map to make coatings blurry, but varying roughness across a surface via a roughness map is impossible.

I now opted for a "stupid" approach for my material library: A low-opacity Nd1000 reflective layer on top of a diffuse base layer, while filling in low-angle reflections via a standard low-Nd coating. It's just as fast as the coating-only approach, makes mappable roughness possible and produces much more convincing results than the additive approach. It is of course by no means physically correct.

Wouldn't it be nice if, instead of applying a coating to a BSDF, one could just add a 'coating layer' to a MXM (like you now add a layer) and inside it, create several 'coating BSDFs', that are weighted together (like BSDFs currently are) and together act as the coating?

[AlexP]

Indeed, it should be fixed ASAP.

[eric nixon]

Gmenzel, its only a problem if you use roughness 0 for the spec, try 0.01... I'm sure they'll fix it.

Also, using coatings for spec is not a good idea, it may appear to work fine in soft lighting but try with bright sunshine + dof and it will be noisy.

Re, Coronas approach, If thats the case (intensity of spec dims diffuse) I'm pretty sure thats wrong. The intensity of the spec at a certain point as seen from a given viewing angle wouldn't dim the diffuse at that point, because the diffuse shading is taking light from all angles... spec reflections are always moving relative to the diffuse surface, as the viewer or object moves, they're independent.

[gmenzel]

Gmenzel, its only a problem if you use roughness 0 for the spec, try 0.01... I'm sure they'll fix it.

Increasing roughness will of course spread the effect out somewhat and make it a bit less obvious. But it can by no means be called a solution. Here the scene with the roughness turned up to 10 (SL16). It still looks like it's glowing. And the noise - well...



This is not a bug. Additive layers work as intended. There is nothing that could be fix here. All this only means that additive layers are not a suitable approach to coated materials.

What Maxwell needs is a coating option with mappable roughness and reflectance. But even falloff maps would be a huge improvement, as they would allow to at least do the whole thing manually, like we did with biased renderers in the past.

[gmenzel]

Also, using coatings for spec is not a good idea, it may appear to work fine in soft lighting but try with bright sunshine + dof and it will be noisy.

True, I noticed that too. With physical sky, colored noise starts to creep in around SL15 (why?), especially outside DOF.

But what else am I supposed to do? As I see it, there aren't any other options. Also, isn't it a rather sad state of affairs when 'using coatings for coated materials is not a good idea'?

If someone from NextLimit could give their thoughts on this issue and maybe provide some outlook on future improvements of the material system, I'd hugely appreciate it!

[eric nixon]

Your just being silly now..., 'coatings' is a function to simulate thin-film interference. The fact that 'coated' materials dont need 'coatings' is just semantic, they're just similar sounding terms.

Additive coated materials in MW3 render really nice, (I'll attach an example) It seems from your tests that there is a problem when using unrealistic settings (which settings are you using btw?), but that can also be seen as an advantage for creative freedom, not something that actually needs fixing. I expect it will be fixed however.

[gmenzel]

Your just being silly now...

I don't think your dismissiveness is warranted. If Maxwell's coating component is really just meant to do thin film coatings, than that would leave an even greater gap in functionality of the material system. But here is what Tom wrote on the matter:

"Additive Blending" is just a mathematical way of summing up reflections and it has *nothing* to do with physical top coats or lacquer layers etc. They are *coatings* and we already have a coating component. "Additive Blending" is nothing else than a computation method.

And according to the manual:

Coatings can also be used to create rough materials with a clear coating. For example, a glossy white plastic can be made combining a diffuse BSDF with a coating.

It seems from your tests that there is a problem when using unrealistic settings

I gave all my settings. Please tell me which parts of them are unrealistic.

Your example doesn't really have much self-reflections, which is where the problem shows.

[eric nixon]

The manual is kind of wrong technically, but its pointing out that option, which works nicely in some situations. Please read Tom's Dos and Donts, posted recently on the maxwell blog. I wasn't aware you posted the settings, but I'm a bit busy to search for them.

Also the bug when using exactly roughness 0, may have been fixed already. I'm not using the latest maxwell version here / haven't tested that issue lately.

Hasn't Tom already acknowledged the issue of self-reflection, and promised that they're working on it? maybe that was another thread? I'm pointing out that the problem is quite minor really and quite easy to avoid, i.e. use a normalized setup for very bright shiny-walls/concave-shapes.

But overall I'm glad your searching for problems.

[gmenzel]

Please read Tom's Dos and Donts, posted recently on the maxwell blog.

The only relevant part of that blog entry would be "[Don't] Use Coatings for creating everyday clearcoats. Plastic approach using 2 BSDF (normal or additive) will be much faster instead.", but this appears to be a recomendation in regards only to speed. Otherwise this would also conflict with what Tom wrote ealier in this thread.

Also the bug when using exactly roughness 0, may have been fixed already.

That seems to be a different issue, though. Increasing roughness does not solve the amplification problem, as I have shown.

Hasn't Tom already acknowledged the issue, and promised that they're working on it?

Not that I'm aware of. He requested my first test scene to investigate, promising to return once he had results. Unfortunately, it seems he didn't get around to it.

I'm pointing out that the problem is quite minor really and quite easy to avoid, i.e. use a normalized setup for very bright shiny-walls/concave-shapes.

Yes, it is possible to work around the problem - as long as you are in full control of the scene, the lighting, as well as other render setting.

But if your objective is to create a library of materials that can be used universally, by anyone, in any scene context and lighting condition, then the problem seems anything but minor.

But overall I'm glad your searching for problems.

Or rather for solutions, I would hope.

[gmenzel]

With all the recent talk about the upcoming V4, I wondered whether there might be any improvements to the material system, especially concerning the issue discussed in this thread. GPU rendering is certainly nice to have. But, at least for me, increased speed isn't the most pressing matter. I have run into the light amplification problem many more times since this thread died and I have now started to use another renderer for certain kinds of projects. I still prefer to use Maxwell where ever I can because of its simplicity and otherwise strictly physical approach.

Since I unfortunatelly wasn't able to get anywhere with this thread, I wonder if perhaps V4 will bring improvements, e.g. a more capable coating component or physical layer blending.

[eric nixon]

Can you post an example? I thought this was fixed already. Would love to hear a devs opinion though.

[gmenzel]

Can you post an example? I thought this was fixed already. Would love to hear a devs opinion though.

There is nothing to simply "fix". If you add reflections on top of an already (even diffusely) reflective surface, you get more light reflected than you should. What is needed is a better approach to physically stacked material layers. Something like the coating component, only more capable.

I have updated all the expired image links in this thread, in case people are interested.