V-ray Next 4.30.02 - Extreme "dancing" white pixels on any material at a distance with glossy reflections

[^Lele^]

Yes, have you read my post above? That's the fix, as of now.
Any -potential- further build would only suppress samples of this kind, not sample them better.

[kosso_olli]

^Lele^ Did you have a chance to look at the two EXR's I uploaded in a previous post?

[kosso_olli]

EDIT: Nevermind, the post is right there before Lele's last reply

[^Lele^]

Thanks for the research and reply Lele, fingers crossed for a working solution in the future.

I've tried both Corona and Arnold, and they both behave exactly the same.

Corona:

Arnold:

The only way to have the issue disappear, currently, is to fix the noise pattern and render to below noise visibility (this is the same 3 frame gif as the rest.):

[kosso_olli]

You are conflating different issues.

a) Teapots file: hide the sun disc.
Do so, the flickering is gone.

I am sorry, but I can not reproduce your results. Please open the attached file and play both frames in Chaos Player. There is still flickering going on, sorry.
Yes, it reduced by a considerable amount, but it is far from gone. There are still flickering highlights in every row of teapots!

https://transfer.px2.de/_4AQ1RQNRgdI4YR

[Vizioen]

Thanks for the research and reply Lele, fingers crossed for a working solution in the future.

[^Lele^]

Here is another test case with teapots along with the scene file.
The blinking is visible on every single teapot.

You are conflating different issues.

a) Teapots file: hide the sun disc.
Do so, the flickering is gone.


b) Your file: the bad geo was creating time-uncorrelated glints. Those disappeared as i hid the sun, and fixed the geo.

the original:

the fixed handle:

subbing the handle for standard geo:


c) Ashley's issue: those are time-correlated, sub-pixel highlights (same as the OP), and that travelling should be fixed by two specific options of ours, but they are currently not working.
That has been escalated to the devs.

[kosso_olli]

Here is another test case with teapots along with the scene file.
The blinking is visible on every single teapot.


Glints_03.zip

origRGB_03_MP4.zip

[Joelaff]

The rest of you post is very valid, but none of this should be used these days. It's not 2005 anymore, and I agree V-Ray should be able to handle this out of the box.

The funny thing is, we never had to do this in LightWave, and LightWave 6 was 64 bit (CPU), 32bit float (images), and worked with HDRI (first app to do so) and terrific GI with about the same or less noise the VRay6 all the way back in 1999. Of course you had no nodes at that point, and had to make your material conserve energy on your own using falloff maps, the animation tools were limited, etc.

Again, this is really VRay's Achilles' Heel right now.

[kosso_olli]

So we are all on the same page, what, specifically, did you use to fix the normals, ^Lele^

Lele welded the coincident vertices and then smoothed the results. This introduces artifacts in other areas, though. In the scene he sent back, the glints did indeed disappear and I was able to reproduce it in the stripped down scene. In the full scene with the enrionment, sadly not. Still do not know why, though.

The rest of you post is very valid, but none of this should be used these days. It's not 2005 anymore, and I agree V-Ray should be able to handle this out of the box.

[kosso_olli]

I actually set the density of the stochastic flakes to 0 because I thought that that was causing the issue

During the tests, a very simple shader was used. A VrayMtl with red diffuse color and a clear coat. So as simple as can be, yet the glints appeared. It happens as soon there are high-gloss surfaces.

[kosso_olli]

I'd be happy to share visual material here, if given Olli's permission.

Sure, you can post them.
One thing I am going to try tomorrow is to merge the parts you sent back into the main scene. For some reason I can not manage to get rid of the glints, whereas I could do that in the stripped down scene with the isolated parts.

[Joelaff]

So we are all on the same page, what, specifically, did you use to fix the normals, ^Lele^

The fireflies, especially with VRaySun, just really need to be fixed on Chaos' end. They are the biggest drawback to VRay right now, and waste a lot of time and effort.

The only way we have found to avoid them is:

Never to use VRaySun for the reflections.

For your reflections use another Standard VRayLight as a disc, often with a radial falloff map. You must NOT set the directionality high. Much over 0.3 and you are apt to get the fireflies again. This light should be as large as you can visually get away with and still have a decent look to the shot. (In other words, if it is tiny you will quite possibly still get fireflies). Start with it much bigger than you think you could get away with, and check and see, accuracy in sun scale is far less important than most people think, and a larger light source typically looks better on reflective surfaces anyway. (Not at all a fan of this trend, especially in cars, to front light them with the sun-- yucko, ugly lighting, and nobody can figure out why their products lack richness... but that is a different discussion on lighting, and, honestly, production budgets that rarely will let you sit out from 9AM to 3PM as you used to when filming because the light was bad.) Luckily, unless you are in space, the sun is never a perfect pinpoint absolutely parallel light source that many people think it is. The sky causes a lot of scattering, and the rays become less and less parallel with more haze, etc.

If you are lighting with an HDRI go into an HDR aware app and make a feathered circular selection around the sun disc in the image and blur it (make the selection big enough that the blur doesn't go into the feather of course). Blur it as much as you can without impacting the look of your shot. Use this HDRI for reflections, and you unblurred HDRI for everything else.

You may also try to blur both the reflection HDRI and the diffuse HDRI, and shrink it as small as you can (scale the image down). This is a tradeoff because the more blurry and smaller you go the less detail you will have, but you can often get away with far less than you think, and this will render much faster as well. Of course with something like a clearcoat on car paint, which demands a very high glossiness, you can't get away with much blurring on the reflections, but you can soften the sun itself as described above.

Sometimes you may use an HDRI where you have blurred the sun quite a bit, and then ALSO add a disc VRayLight which you line up with the sun. This can render a lot faster in many cases, and you can use one (significantly blurred sun) HDRI for everything, along with the VRayLight.

When doing these things do whatever you need to do keep the shadows at the correct sharpness level. (Separate passes, or separate lights, etc.)

Of course, we shouldn't have to do any of these things, as they can literally eat hours of production, especially if you have to render animations at near full quality every time when you are setting them up to make sure there are no fireflies.

[Vizioen]

These are quad modelled cars, downloaded from 3Dsky. But I have this with about any car I purchased from either 3DSky, turbosquid, evermotion, etc.

'm not an automative artist but I need to use cars a lot in my visuals for logistic projects.

I actually set the density of the stochastic flakes to 0 because I thought that that was causing the issue at first but it never fixed it. I've uploaded some of the cars and I'll send you the link through mail. Thanks for taking a look. I hope it's not something I did wrong now though . I didn't save the scene but it's easily reproducible with just a vray sun and a static camera (doesn't matter if the sky is in a dome light or not).

[^Lele^]

Well there's definitely something weird going on. I tested a similar setup because I also have problems (mostly with cars) with dancing highlights. So I made a test setup with 4 cars, that should have more or less proper geometry. All settings at default for animation, nothing is animated even (animated camera or whatnot has same results though). Shouldn't all the frames be the same result?

Granted it's a small resolution: (1000x700):

(20 frames in a loop)


But a bigger resolution (2000x1400 but scaled back to 1000x700 because the forum can't handle files bigger than 10MB ) produces similar results too, albeit less bad:



The weird thing is it even flickers in the glass at lower resolution, I always thought it was just the carpaintMTL.

Are the cars coming from CAD, or are they quad modelled?
The issue does *not* appear, f.e. in Olli's scene, with max-created geo of analogous shape or dimension in place of the original geo.
Nor did it happen again for me -for the three pieces I fixed-, I can't fathom what the rest of the car, or any other piece of CAD geo on the planet, does.
I can surely try your scene out though. Ashley, and we may find pointers.

On the shading side, there are surely a number of ways to create glints: low sampling of stochastic flakes will do that, f.e.. or a visible sun disc.
But that was never the case for Olli's scene, as far as the pieces i was given, and thew result i managed to produce.
I'd be happy to share visual material here, if given Olli's permission.