Material reflections & max sample intesity

[cgrobo]

I ran into a strange issue with a completely non reflective material. The dark ball has both reflection and reflection glossiness set to 0, yet it seems to be receiving light from the chrome material (red arrows). Setting the max sample intesinty to 1 seems to get rid of this issue, however I assume changing this value to 1 in other rendering situations won't work.


The effect is even more pronounced with a glossier material/ball (pic 44, 55 and 66)


Why is this happening?

[pokoy]

I think these are GI caustics, sort of non-sharp caustics coming 'for free' with the rendering method. You can probably get rid of these when you use a rayswitch material and use a non-reflective material in the GI slot (or both reflected and GI).

[Nejc Kilar]

Yep, this to me looks like the usual and expected reflection behavior.

When the surface is not glossy it will bounce rays all over the place basically whereas with the glossy material they'll be more focused and that will make a bigger, more visible impact on things. When you limit MSI you're basically saying "hey, those really strong rays should be clipped" and as such you're then not getting them to come though.

[cgrobo]

Appreciate the replies. How come this "expected behaviour" doesn't occur in Fstorm? If it does, it doesn't look smudgy. Attaching images.

[pokoy]

I may be wrong but the metal needs to be less blurry and probably needs the same amount and rotation/angle of anisotropy (which will play a role in how the light is reflected on the surface).
Thinking of anisotropy, this is probably contributing to how 'smudgy' it looks. But the reflecting mirror material should be as close as possible to compare.

That said, BRDF models probably aren't the same for both renderers, so differences are to be expected. Again, you might want to try the Rayswitch mtl wrapper if you run out of patience.

[cgrobo]

I may be wrong but the metal needs to be less blurry and probably needs the same amount and rotation/angle of anisotropy (which will play a role in how the light is reflected on the surface).
Thinking of anisotropy, this is probably contributing to how 'smudgy' it looks. But the reflecting mirror material should be as close as possible to compare.

That said, BRDF models probably aren't the same for both renderers, so differences are to be expected. Again, you might want to try the Rayswitch mtl wrapper if you run out of patience.

Attaching two comparisons: metal gloss set to 0,7 and to 1 in both corona and fstorm. Ball gloss 0,5. Anisotropy off. Corona max intensity set to default 20. In Corona there are visible "reflection smudges" at 0.7 gloss but not at 1 gloss. While in fstorm no smudges at all.

[cgrobo]

Metal gloss 0,7. Ball reflection off.

[pokoy]

Can you try a legacy Corona material on the black sphere? Assuming you are using Physical now...

Otherwise I'd suggest to provide the scene for devs to reproduce - maybe there's really something going on...

[cgrobo]

All comparisons were done with Corona Legacy. But here is a comparison between c legacy and c physical. Metal gloss 0,7. Ball gloss 0,5. Smudges maybe look even worse with physical.

[pokoy]

I agree this doesn't look too great... and not expected either, even with anisotropy etc it doesn't look natural.

[Nejc Kilar]

Yeah that does look potentially funky so thank you a lot for bringing this up. These last examples are imho very useful as well. We'll discuss it internally and we'll post a reply here once we know more.

[pokoy]

@OP - Would it be possible for you to share the scene, I'd like to see for myself and having the exact same file would be the best starting point.

[maru]

Hi, here are my initial observations. They may not be 100% accurate. :)

Post 1 - Reflective caustics (HDRI lighting falls onto the metal, then this is reflected onto the non-reflective surface). You see "splotches" because various shapes from the HDRI are producing various intensities on the surface. It goes away when MSI is lowered, because then light bounce intensity is capped.

Post 2 - We don't know how fstorm works internally, but it is possible that it uses heavy clamping or does not produce GI from glossy reflections (these are just guesses). So it is missing the effect that happens in real life and is visible in Corona.

Post 3 - Exactly the same comment as to Post 2.

Post 4 - Same.

Post 5 - https://forum.corona-renderer.com/index.php?topic=41005.msg217107#msg217107 - this one is interesting. I have two questions about it:
- Are the inconsistently splotchy patterns reproducible if you switch from Path Tracing + UHD Cache to Path Tracing + Path Tracing?
- Are you using some kind of denoising?

I am asking because both the UHD Cache and denoising can produce varying splotchiness. The UHD Cache takes care of GI and also affects glossy surfaces, so it may be related. Switching to PT+PT would remove UHD from the equation.

Update: it could also make sense to add CESSENTIAL render elements and see where those splotches appear.

Update 2: attaching a similar scene from Max 2024 and 3 renders (no changes, just hit render 3 times). It's a non-reflective ball and a reflective floor + 3 Corona lights with high directionality. You can see the changing pattern. This is caused by the UHD Cache "splotches". The solution would be to switch to PT+PT, or possibly the "animation" preset of UHD Cache may help.

[cgrobo]

Thanks for the replies. So maru got it right, UHD Cache is the culprit. Setting to PT+PT resolves this issue. Attaching the image.

I think Fstorm uses only PT. But even when I compare corona PT render with metal gloss set to 1, to fstorm, the reflection doesn't look right (3rd and 4th image)? There is a big reflection at 1 oclock that just disappears on the left, while in Fstrom it is evenly distributed/reflected. Is this because of maru's "post 1"?

Can this UHD funkiness be improved?

[cgrobo]

Comparison with HDRI rotated. PT enabled.

[brr]

Can this UHD funkiness be improved?

Hi,
did you tried 4k cache as well? In most of the cases it works really well and gives better results than UHD cache.

[cgrobo]

4k cache removes this problem. I was under the impression that 4k cache should be used for animations. But reading about it now, it seems like it has replaced UHD cache. Thanks for the tip.

[maru]

The entire point of the UHD Cache is that it greatly speeds up rendering at the cost of some bias. Usually, a bit of splotchiness is unnoticeable, unless you render a sequence, but that's why we have the "animation" preset and other UHD options: https://support.chaos.com/hc/en-us/articles/4528617365649

It is possible that the 4K Cache is better in terms of those tiny splotches, but please keep in mind that it's still experimental and may sometimes produce unwanted results (we would really appreciate reports :) ).

[pokoy]

I'm happy Maru was able to identify this, I always assumed GI caches wouldn't have any effect on reflections but it totally makes sense. Good to know for the future.

[brr]

It is possible that the 4K Cache is better in terms of those tiny splotches, but please keep in mind that it's still experimental and may sometimes produce unwanted results (we would really appreciate reports :) ).

Hi Maru,
I have been using the 4k cache by default since its release. From my experience, it not only provides more stable GI without any splotches but also results in faster pre-calculation and more rays per second during rendering. It works really well, and in my case, I encountered an issue only once while rendering an animation, where about 3 frames out of a 1500-frame sequence were "broken." I simply re-rendered them, and it turned out fine.

However, it's interesting to note that texture baking is not an option with the 4k cache; it produces a lot of splotches everywhere, especially if the model has explicit normals.

Sorry for going off-topic. I hope that the 4k cache will go final sometime.
If you need some scenes with texture baking where I had really visible problems with 4k cache, then I can provide them to you next week.

Best regards