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

[Spatial]

800% zoom in showing issue:




This one has plagued our studio for years and we've always solved it just by turning off v-ray sun specular.

This primarily happens with our car materials and any material that has a glossy value above 0.8 or more where the meshes are quite some distance from the camera in sub-pixel land.

I have methodically tried every single v-ray setting under the hood (have searched extensively through the forums) and nothing prevents these bright white pixels from occurring except for two things:

1.) turning off sun specular (sun is always set to invisible by the way)
2.) turning off or lowering gloss or reflection amounts on materials. (even this can just cause softer dancing hotspots)

Annoyingly, I installed Corona as a test and it has a simple highlight compression setting that when set to 1.0 instantly solved the problem!

I cannot find the equivalent setting in v-ray if it exists.
I have tried every combination of color mapping, clamping, burn values, sub pixel mapping, color mapping types...and on and on...literally cranking, lowering, checking or unchecking every single setting possible in v-ray

I have uploaded the above simple test scene here:
https://tinyurl.com/y5mbw4qc

Any insight would be greatly appreciated.

Thanks,
KG

[hermit.crab]

This topic has some tips about similar issues. Another thing that significantly reduces the fireflies is increasing the Sun Disc size through the Size multiplier ( a high value like 100 seems to eliminate them completely on your scene without a drastic lighting change).

[Spatial]

Thanks for the reply Aleksandar.

Sun size to 100 creates ultra soft shadows so it's not a tenable solution.
Unless you have another hack that allows a massive sun size but retains sharp shadows?

The link you provided has the same old suggestions that I outlined in my initial post and I have never been able to get them to work.
If you can get those tips to work on the test scene I provided, I'll be extremely happy to be wrong.

Hopefully support can take a longer look as this happens to us in nearly every single project and it has been very frustrating for a very long time.

Corona nailed it with its highlight suppression setting. Is v-ray just not capable of handling this?

[Joelaff]

Thanks for the reply Aleksandar.

Sun size to 100 creates ultra soft shadows so it's not a tenable solution.
Unless you have another hack that allows a massive sun size but retains sharp shadows?

The link you provided has the same old suggestions that I outlined in my initial post and I have never been able to get them to work.
If you can get those tips to work on the test scene I provided, I'll be extremely happy to be wrong.

Hopefully support can take a longer look as this happens to us in nearly every single project and it has been very frustrating for a very long time.

Corona nailed it with its highlight suppression setting. Is v-ray just not capable of handling this?

You could use two VRaySuns. One affecting only specular which is large. The other affecting everything else which is small.

Also, depending on the severity this may help or not, but it makes the best dang looking tiny specular highlights, and that is to use the Soften Image filter around 2.98-3.98.

Also increasing minimum subdivs (2-16) usually helps tremendously.

[Spatial]

Hi Joelaff, great suggestions. We had considered using 2 suns moving forward but was hoping for v-ray to just...work. To simplify procedures for all staff. You know what its like, someone updates the traffic xref because of a client comment right before renders and in the rush the new cars don't get added to the include and exclude lists in the suns. Nightmare.

I had also considered creating distance blends for glossy materials that reduces the gloss amount by x when the camera was y distance away, but this is a bit of a library overhaul and who has time for that.

The 2 sun hack might have to be the way to go for now until V-ray can catch up to Corona though :/

Maybe our studio which outputs a huge number of high fly through videos in giant infrastructure projects tends to get this hot pixels issue more frequently than others who tend to do closer, smaller scenes. I don't know. But we're usually working in many km's of project space and usually flying high above which is when this problem presents itself the most.

I have definitely tried cranking subdivs. I even went up to 24/24 and the giant 4 pixel square white hot highlights were not subdued even a little bit.
on a side note 2/16 is too much overhead on our average 3000 to 5000 frame outputs. That would get very expensive very fast.

I will try the soften image filter and see how that goes! thanks for the suggestion!
Hopefully a more permanent solution under the hood is on the cards. Talk to the Corona devs! They fixed it

[^Lele^]

Your AA at 1-6 is way too low.
Your LC subdivs should be at least 3000, with a retrace of at least 8.

Reset the settings to their default, select the bucket sampler, set min AA to 3 or more (so it'll find the detail), select the LC preset named "Animation" (or follow the guidelines above), and render with motion blur to get the correct result.

V-Ray is doing it right, in this case.
The longer the blur length, the smoother the transitions of highlights across the discontinuous geometry will look.
If you render without motion blur, there is no correlation between frames (and your camera is trucking in, ensuring angles change at each frame, for all pixels.), and with a highly discontinuous surface like the bonnet and various glasses, this behaviour is wholly expected, and entirely correct.

If you prefer to kill the highlights, a-la Corona, feel free to turn on SPM and lower burn to some very low value.
It will not help here, however.

The reason has to do with shaders setup: make those glasses (Multi-sub, ID 5) something like 0.8 gloss: at this distance no difference will be perceivable anyway, but thew broader gloss tracing cone will smooth the transitions of the highlight.
Likewise, there is no need to set a reflection color to white.
The same applies to the headlights reflectors and glasses (those too catch a glint. again, it's the correct thing to do for V-Ray.), ensure they aren't total reflectors (no fresnel, no gloss in yours), or once the angle is right, they'll shoot the sunlight back at you (think a hung mirror turning in the wind, far in the distance. That's what's happening here.)

Attached, four videos (not denoised):
a) a video of the sequence rendered like suggested above,with motion blur at defaults, but without any clamping of the highlights or changing shaders.
b) No motion blur, but with a Reinhard burn of 0.01 and SPM activated.
Both still exhibit discontinuities.
c) I lowered gloss marginally on the windows glasses, and on the headlights chrome reflectors and glasses for the blue car (I haven't changed all the shaders accordingly, so something still glints. I'm sure you get the gist.).
You'll notice that is the most coherent so far.
d) Lowered gloss for shaders and motion blur with a length of 1 frame. Glinting is reduced to nothing at all for the shaders i changed.
I didn't compress highlights for the last two videos.

As a side note, i am *very* interested in studying the Corona version of this scene which managed frame-by-frame continuity by simply rolling the highlights intensity back.
I'd be grateful if you shared that one as well.

[Ali Essawy]

Maybe you could try to decrease the Max ray intensity with subpixel mapping checked or with the clamp output to 0.95, it will darken the scene a little bit but it will reduce the hot spots

[Joelaff]

Maybe you could try to decrease the Max ray intensity with subpixel mapping checked or with the clamp output to 0.95, it will darken the scene a little bit but it will reduce the hot spots

Is this what corona is doing? Or do they do something different?

[Spatial]

Thanks for the response Ali, unfortunately i have already tested those values and they do not change anything with this issue.

Wow Lele, thanks for such a detailed response with test cases and videos! I really appreciate the help.
AA makes no difference to this issue. I've cranked min and max subdivs to astronomical levels as well as increased LC subdivs to very little effect.

Lowering burn has a far different effect to Corona's fix which appears to leave the image completely intact while only removing hot pixels. It's a bit of magic I wish v-ray had. This whole thing also keeps us away from glare and bloom which only makes this problem even more in your face.

I know reducing gloss and reflection helps this issue but that's a huge task. We also have quite a few forest pack packs that exhibit this behavior with leaf spec and gloss and its prohibitive to go changing all those libraries too.

As far as I can tell after years of dealing with this one, is that while v-ray is being "entirely correct" it's mostly being in the category of entirely unwanted and completely undesirable which is the last thing a busy studio needs.


As requested, here is the Corona test case on the same scene.

01.) - Corona also renders it the same "correct" way by default with sparkly highlights




02.) - Highlight clamping value in the system settings



03.) - Clean as a whistle



These are 15 second frames showing that high AA settings are not required, plus i didn't have to go changing all the materials in our prop libraries

Corona Scene: Onedrive link

Maybe switching to Corona is the answer but that's a whole other kettle of fish and would be like trying to turn the titanic during a busy production schedule.
I just hope Chaos can do something similar to Corona instead of asking the customer to change years worth of materials and libraries.

The only quick fix at this stage seems to be the 2 sun method(One sized at 100 but only affecting specular contribution) + some attention to detail in scene management.
Even if v-ray sun had some sort of option to disable this flickery pin-point specular behaviour or something. Or a mega v-ray sun that retains hard shadows by default haha.

Thanks for all the help so far!

[^Lele^]

Eh, it entirely kills highlights, basically turning your shaders to diffuse.
If you like this, your best bet is to wait for our version of a similar tech (there should be a build lurking somewhere in the max 5 beta forums.).
I would stay away from this like if it was the plague, because those highlights *need* to be there, in one form or the other, to marry those parts with the rest of the scene, but i can see the usefulness if you're pressed for time.

[Spatial]

That is an incorrect statement.

Its just clamping the highlights from peaking out way past acceptable ranges before it gets to the frame buffer.
I just read up a little Corona documentation and my theory was right about that.

The highlights are still there. They're just not insane

I've since done a range of tests with highlight clamping values. 1.0 is pretty extreme in the first test. (still infinitely better than the alternative and close-ups of materials still look beautiful with just the smallest of pixel highlights being clamped out:

Example: (middle click to compare in browser tabs)
Highlight clamping = 0
Highlight clamping = 1

It gets really interesting when you set the highlight clamping value quite high.

Check out what a value of 100 does:



This is it! That's the holy grail right there. A visible specular highlight that is smooth and looks great. No crazy pixelly dancing firefly shenanigans or white hot pixels that flicker on and off constantly. No insanely high settings and long render times required to fix it. Just works.

Avoiding this like the plague is a bit silly. This option is in Corona to fix a very long standing problem and it does it spectacularly.

Please let this be the same tech you are talking about in the nightlies and not the "highlight killer" you are describing.

side note: do I need to actually make a formal support request or are support actually looking at this?

[^Lele^]

The statement is correct if you set clamping to 0, ofc.
EDIT: not even. 0 is a special value which renders unbiased, reading the manual. But it's clearly not so. Things have changed from what's written in the helpdesk.
I may sound testy, but i always flag stuff, even if i personally don't like the approach, to the relevant people (which ain't support, here.).
You are being read.

[Spatial]

Yeah 0 is the default which = highlight clamping off.
The higher you go you eventually reach fully unclamped again though, so somewhere in the middle gives a really sweet result with no artifacts. It's a beautiful thing.

Cheers Lele, I might put a ticket in anyway. Hoping the forum would help, but this is just something v-ray needs to be better at under the hood I think.

[^Lele^]

Sure, open a ticket if you want, i'm not gatekeeping, i'm merely helping to sort out the amount of data the forum produces for busy coders.
Either way, this is no news, work's been going on on this for a long while, now.

[^Lele^]

This is rendered from the original file, without any modification (besides turning off denoising, and setting the output to an EXR), with the fireflies suppressor build mentioned before.
Renders were below 5 secs per frame.

This is the internal build, should you want to try it, for max 2020.
A few caveats apply:
a) This is from well before the release of v5. It should be used *exclusively* to try out the firefly suppression qualities in scenes which were problematic.
b) It *requires the bucket sampler*. And more than one AA Sample (as usual, defaults will serve you well.)
c) It should be representative of speed for the firefly suppressor only.
d) The feature is always on in the build, if you use buckets. Install in place of the old one, switch to bucket sampler, render away.
To compare render times and looks, simply switch to the progressive sampler.

If a build is needed for another max version, Vlado would be kind enough to make you one.