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

[kosso_olli]

Here is the test with the specular disabled for the sun and an additional disc light far away.
It is fiddely to get the right position and appearance, but the render is clean.
However, if V-Ray could handle these situations out of the box, I would be very happy.

[^Lele^]

Sadly, a min subdivs of 16 makes no difference to the teapot scene above.
And it's a min amount that's not feasible for the kind of work Olli does: rendering most of his cars as fixed with high sampling would take forever.

[Morbid Angel]

Kosso_olli - what are your sample settings?

[kosso_olli]

I will try to disable specular contribution for the sun, and then placing a directional disc light to care for the specular. Hope it get's better this way...

[Joelaff]

setting the min subdivs to 10-15 should fix it. Honestly though you have to understand that sampling a bright dot from an hdr is no easy task for a raytracer.

No doubt, which is why people often blur the HDR for such things.

Objects also glint in real world too the further they are.

True, but, sadly, that flickering almost never looks like what you get from VRay (or many raytracers), especially like the example above. The glinting in the real world is much broader, and hardly ever changes that much from frame to frame. In a slow moving render averaging frames together in post before applying some sort of bloom can help.

[Morbid Angel]

Lele, it seems there is some misunderstanding.

In the teapot file from post #59 the camera is moving.
In your file you uploaded in your last post, it is not. We know how to render consecutive frames with static camera without getting randomly placed highlights by now. The solution was proposed by Joelaff in post '33 by deactivating locked noise pattern:



His solution was confirmed by me in post #34, with an exception:



So, I opened the scene you uploaded today, animated the camera ever so slightly, and voila: Dancing highlights (i just set metalness back to 0, sun disc is disabled of couse):




If you could take a close look at that scene again to find a working solution, that would be great. For over a month now, I am not able to produce a sequence that is looking clean. I'm seriously running out of ideas. File is attached. Thank you so much!

teapots_olli_01.zip

setting the min subdivs to 10-15 should fix it. Honestly though you have to understand that sampling a bright dot from an hdr is no easy task for a raytracer. Objects also glint in real world too the further they are.

[^Lele^]

Oh, sorry by "sufficient" I meant capable of maintaining the desired "look" of being in the sunlight, while simultaneously ameliorating the flickering.

Eh, yes, ofc the sund disk's glints will be gone.
We could also call them tiny speculars, and then we wouldn't be fidgeting with nomenclature, ahah

Also, making the VRaySun bigger (I am assuming this is what you meant) still keeps full directionality/parallel rays. These are more accurate, but matter very little in the speculars, and lead to more flickering (hence my suggestion NOT to set a high directionality.

Making the v-ray sun bigger makes the shadows softer, and ofc the reflection of the disk is bigger too.
However, the kind of sizes one would need to make the sun disk easy to sample would also make the render awfully slow (besides being utterly unrealistic.).
This is why the procedural sky allows for a custom sun node: plausible values aren't always friendly.

Of course as kosso_olli keeps saying, we shouldn't have to do any of this. It should "just work."

As mentioned, the current "state of the art" is suppression (f.e. Corona): so your highlights are going to be gone, which is highly problematic a behaviour in itself.
If you know something the V-Ray, Arnold, Corona and Renderman Devs don't about sampling a sun disk more reliably, please get in touch: there's serious money to be made.

Here's how big a sun disk is (while being in the several millions float for intensity).

[Joelaff]

I haven't found one where it doesn't remove the glints.

Oh, sorry by "sufficient" I meant capable of maintaining the desired "look" of being in the sunlight, while simultaneously ameliorating the flickering.

Indeed, studio work is generally a better option than using the sun directly, just like it often is IRL.
Your approach is also a wee bit better than simply making the sun disk bigger, as that would impact shadows and sky as well.

Also, making the VRaySun bigger (I am assuming this is what you meant) still keeps full directionality/parallel rays. These are more accurate, but matter very little in the speculars, and lead to more flickering (hence my suggestion NOT to set a high directionality.

Of course as kosso_olli keeps saying, we shouldn't have to do any of this. It should "just work."

[^Lele^]

Not sure that is sufficient in a lot of cases.

I haven't found one where it doesn't remove the glints.

However, you can augment the highlights as I stated before, with an additional VRayLight disc setup to mimic the sun's position (and without setting the Directionality too high), making this light as large as you can without sacrificing the look (bigger will flicker less and render faster).

Indeed, studio work is generally a better option than using the sun directly, just like it often is IRL.
Your approach is also a wee bit better than simply making the sun disk bigger, as that would impact shadows and sky as well.

[Joelaff]

so zero the sun's contribution to speculars, and you'll be fine (your highlights will come from the sky, and the sun's halo.).

Not sure that is sufficient in a lot of cases. However, you can augment the highlights as I stated before, with an additional VRayLight disc setup to mimic the sun's position (and without setting the Directionality too high), making this light as large as you can without sacrificing the look (bigger will flicker less and render faster).

[^Lele^]

They are not randomly placed at all, they are well time-correlated.
The angle gets just too shallow from one frame to the next, the sun is partially hit, the highlight drowned by the samples that missed it.

This is the same issue as the OP, for which the only possible fix is to kill those paths.
The issue however is unsolved as of yet (and while it's actively researched, there is no ETA yet.), so zero the sun's contribution to speculars, and you'll be fine (your highlights will come from the sky, and the sun's halo.).

[kosso_olli]

I can't fathom what you did: it works perfectly fine for me (3 frame gif below + max 2024 scene.).

Lele, it seems there is some misunderstanding.

In the teapot file from post #59 the camera is moving.
In your file you uploaded in your last post, it is not. We know how to render consecutive frames with static camera without getting randomly placed highlights by now. The solution was proposed by Joelaff in post '33 by deactivating locked noise pattern:

I wonder if Lock Noise Pattern would at least keep them consistent.

His solution was confirmed by me in post #34, with an exception:

Just had to check this, because I did not think of it: Indeed it does!

EDIT: It does only work if the camera is static. As soon as it moved, you get randomly placed hightlights again.

So, I opened the scene you uploaded today, animated the camera ever so slightly, and voila: Dancing highlights (i just set metalness back to 0, sun disc is disabled of couse):




If you could take a close look at that scene again to find a working solution, that would be great. For over a month now, I am not able to produce a sequence that is looking clean. I'm seriously running out of ideas. File is attached. Thank you so much!

teapots_olli_01.zip

[^Lele^]

Sorry, which post do you mean?
The two EXR's in that file already had the sun disc hidden. It works for some of the teapots, but not all of them.
Locking the noise pattern and setting a low noise threshold did not help either.

I can't fathom what you did: it works perfectly fine for me (3 frame gif below + max 2024 scene.).

[kosso_olli]

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.

Sorry, which post do you mean?
The two EXR's in that file already had the sun disc hidden. It works for some of the teapots, but not all of them.
Locking the noise pattern and setting a low noise threshold did not help either.

[Joelaff]

I figured you just toss or clamp samples outside some threshold of standard deviation.

I am sure there are more intelligent approaches, like adding new samples in the cases outside the threshold, or blending in the smoothed light cache solution for such pixels, though maybe such methods are simply too slow.

I hadn’t seen this newer technique. Fascinating problem, really…

https://diglib.eg.org/bitstream/hand...96/121-132.pdf