PDA

View Full Version : Lightwave 2018 fireflies



Tim Parsons
02-22-2018, 12:58 PM
I've been using LW2018 since release and I think it's pretty darn good all the way around. But the one thing that is killing me is the fireflies! I've seen all the videos and read through every piece of documentation and they still exist. The only way I have been able to reduce them to reasonable levels is to increase the reflection samples to 32 plus samples. Adaptive Sampling is worthless. We need a fix! Other renders have solved this issue and NT needs to as well. Other than those fireflies the renderer is gorgeous! And it's really quick to setup great looking materials. Good job NT! Now fix the fireflies. :)

Sensei
02-22-2018, 01:12 PM
Did you try Limit HDRI.. ?

Tim Parsons
02-22-2018, 02:23 PM
Did you try Limit HDRI.. ?

Yes - and that helps but generally to me it seems it just breaks them up into smaller less bright specks. I don't understand why AS wouldn't get rid of them. If it is looking at each pixel and compares it to surrounding pixels wouldn't it be able to determine that - yes I need to attack this super bright pixel! I can turn up the filter radius but that just makes a blurry mess.

Sensei
02-22-2018, 02:27 PM
I know what should work: double (or triple) render resolution, and halve it in Photoshop.. ;)

Tim Parsons
02-22-2018, 02:30 PM
Yeah - that's a thought. Although my client wants final output of 4992 x 4992!

gar26lw
02-22-2018, 04:55 PM
it would be handy to setup an example scene with a sphere, box etc, submit a bug report and then post it here and see if the community can come up with a solution.

vonpietro
02-22-2018, 05:51 PM
Increasing ray reflections took care of alot of the problems - metals are very prone to fireflys

gar26lw
02-23-2018, 01:10 AM
Increasing ray reflections took care of alot of the problems - metals are very prone to fireflys

are you using conductor or pBRDF ?

vonpietro
02-23-2018, 01:56 AM
both - although it works better for conductor

Revanto
02-23-2018, 04:00 AM
Yeah - that's a thought. Although my client wants final output of 4992 x 4992!

There used to be a plugin for LW that allowed you to render out in large resolution strips and then compile them back in Photoshop for a final render. It's not impossible to set something up like this yourself. However, this is best for still image renders and not for movies (unless you have time and/or patience.)

Yeah, the fireflies suck....

Cheers,
Revanto :p

Tim Parsons
02-23-2018, 06:17 AM
Increasing ray reflections took care of alot of the problems - metals are very prone to fireflys

Yes - increasing reflection samples to a high setting (32 or so) takes care of them for the most part but kills render times. At 5K, renders are in the neighborhood of 4 hrs. I am working with a wood surface/material and I do have a sleight metalness setting to make it pop, so I need to look at that. :)

MarcusM
02-23-2018, 06:56 AM
Noise filter in Specular_Indirect buffer can help. But I can't see way to connect this in final render after use noise filter in one buffer, directly in LW...
140275

raymondtrace
02-23-2018, 07:48 AM
...Other renders have solved this issue...

They have? It still seems to be a matter of wrangling settings.

https://www.blenderguru.com/articles/7-ways-get-rid-fireflies

jwiede
02-23-2018, 05:07 PM
They have? It still seems to be a matter of wrangling settings.

https://www.blenderguru.com/articles/7-ways-get-rid-fireflies

Dated 2014. Cycles is quite a bit more well-behaved now.

Devs needs to address the LW2018 inability to apply noise reduction to individual passes and recompose them into final render automatically, for starters. Users shouldn't have to manually recompose the final render by hand from scratch in a compositor, just to apply noise reduction to specific passes.

The devs also need to better tune the engine itself so it's less prone to producing such artifacts in the first place. Other commercial path-tracer engines seem to have largely solved this issue, so it CAN be mitigated/solved.

erikals
02-24-2018, 08:50 AM
https://www.blenderguru.com/articles/7-ways-get-rid-fireflies
nice link.

in regards to techniques for reducing noise, i'm sure they are working on it.

gerardstrada
02-27-2018, 04:41 PM
Another idea which solves the problem without losing dynamic range:

http://forums.newtek.com/showthread.php?155992-This-has-got-to-be-a-render-bug&p=1536731&viewfull=1#post1536731



Gerardo

Axis3d
03-02-2018, 07:18 PM
I'm liking the new render quality from the new renderer, but the grain and the fireflies are killing me.

140469

erikals
03-02-2018, 08:26 PM
did you do this ?
http://forums.newtek.com/showthread.php?156280-Really-NewTek/page6

gar26lw
03-02-2018, 08:48 PM
selectively jack samples higher.

Snosrap
03-02-2018, 09:38 PM
I'm liking the new render quality from the new renderer, but the grain and the fireflies are killing me.

140469

The thing that gets me is that there is no reasonable reason for those specks to show up in most of those locations in your render. So random. That is why in my opinion its a bug.

Asticles
03-03-2018, 05:59 AM
Usually if you use an overbright light source, there will be fireflies, as in other renderers.
Please check your light sources, just in case.

Disable ones to isolate the source of the fireflies.

Regards.

Axis3d
03-03-2018, 08:38 AM
did you do this ?
http://forums.newtek.com/showthread.php?156280-Really-NewTek/page6

Yes. I did not model the vehicles, though. Modeled and textured everything else, plus the render.

Axis3d
03-03-2018, 08:47 AM
Usually if you use an overbright light source, there will be fireflies, as in other renderers.
Please check your light sources, just in case.

Disable ones to isolate the source of the fireflies.

Regards.

The area that is the grainiest (the ticket booth lights have bright, luminous surfaces), is right inside the front lobby of the theater. There are other luminous polygons on the light bulbs of the marquee itself.

Other than that, I am only using two spherical lights - one in each of the street lamp posts.

Axis3d
03-03-2018, 08:50 AM
did you do this ?
http://forums.newtek.com/showthread.php?156280-Really-NewTek/page6

I will check that out in a bit, erikals, thanks. It's gonna take some time to learn the new ways to optimize renders with the new PBR.

Asticles
03-03-2018, 10:32 AM
Edited.

Axis3d
03-03-2018, 12:33 PM
Axis3D, just one question, what Render tile size do you have?

In one render I have it seems that 32 gives me less fireflies, maybe it is scene based. Just a first impression. It shouldn't happen.

Regards

I am using the default of 32. I have done a few tests changing it to different settings, with no change in quality or time. So I just leave it at 32.

The grainiest part of my render was in the Diffuse Indirect channel, which only seems to be cleaned up by more camera samples. But those random, single-pixel fireflies are just so weird. If I change one setting, only slightly, they may show up in a different part of the render. 99.99 percent of the rendered frame looks fine except for these random fireflies. I wish there was some threshold setting made just for these which could just average those pixels out.

It definitely seems to deal with the roughness setting from what I'm seeing. If I turn the roughness to zero, they disappear. But surfaces tend to need roughness. It seems strange that the last release of LW (2018.0.2) had a significant change in the roughness setting from the previous version. I wonder if they are trying to fix the roughness settings.

Anyway, here is the next version with less grain, but still fireflies:
140485

MarcusM
03-03-2018, 01:16 PM
I am using the default of 32. I have done a few tests changing it to different settings, with no change in quality or time. So I just leave it at 32.

The grainiest part of my render was in the Diffuse Indirect channel, which only seems to be cleaned up by more camera samples. But those random, single-pixel fireflies are just so weird. If I change one setting, only slightly, they may show up in a different part of the render. 99.99 percent of the rendered frame looks fine except for these random fireflies. I wish there was some threshold setting made just for these which could just average those pixels out.

It definitely seems to deal with the roughness setting from what I'm seeing. If I turn the roughness to zero, they disappear. But surfaces tend to need roughness. It seems strange that the last release of LW (2018.0.2) had a significant change in the roughness setting from the previous version. I wonder if they are trying to fix the roughness settings.

Anyway, here is the next version with less grain, but still fireflies:
140485

This fireflies can be very easy "corrected" by noise filter in diffuse indirect buffer. But as we know, now we cant see result in final render after use noise filter on any buffer (directly in LW).
You have turned off sample backdrop in GI?

Axis3d
03-03-2018, 01:24 PM
This fireflies can be very easy "corrected" by noise filter in diffuse indirect buffer. But as we know, now we cant see result in final render after use noise filter on any buffer (directly in LW).
You have turned off sample backdrop in GI?

Sample Backdrop has been on for these renders. Turning it off drastically changes the look of the lighting in the scene.

Also, here's a quote from the online help file about the Noise Filter:

"Noise Filter Options

This system, new to 2018, allows you to control a bilateral filtering of your render to give a smoother effect without needing endless anti-aliasing. Meant to remove fine grain that would take too long for the renderer to deal with, it won't work effectively on high contrast grain."


Unfortunately, fireflies are high-contrast grain.

Chris S. (Fez)
03-03-2018, 02:38 PM
I believe you would benefit from Rebel Hill's LW 2018 rendering/surfacing video training. Sampling the backdrop is not the optimal method for 2018. He goes step by step to explain best methods to setup both interior and outdoor scenes....and hybrid scenes. It will save you a ton of time and experimentation.

Morgan Nilsson
03-03-2018, 02:47 PM
I am using the default of 32. I have done a few tests changing it to different settings, with no change in quality or time. So I just leave it at 32.

The grainiest part of my render was in the Diffuse Indirect channel, which only seems to be cleaned up by more camera samples. But those random, single-pixel fireflies are just so weird. If I change one setting, only slightly, they may show up in a different part of the render. 99.99 percent of the rendered frame looks fine except for these random fireflies. I wish there was some threshold setting made just for these which could just average those pixels out.

It definitely seems to deal with the roughness setting from what I'm seeing. If I turn the roughness to zero, they disappear. But surfaces tend to need roughness. It seems strange that the last release of LW (2018.0.2) had a significant change in the roughness setting from the previous version. I wonder if they are trying to fix the roughness settings.

Anyway, here is the next version with less grain, but still fireflies:
140485

What are your lighting settings? My first guess would be that you are sampling the backdrop from multiple sources which is causing too high contribution to indirect specularity.

If using environment light, don't sample backdrop in GI, same goes if you use portal lights. At the same time, if you are not sampling the backdrop in GI switch the shaders to raytrace reflection and refraction only.

Axis3d
03-03-2018, 03:06 PM
I believe you would benefit from Rebel Hill's LW 2018 rendering/surfacing video training. Sampling the backdrop is not the optimal method for 2018. He goes step by step to explain best methods to setup both interior and outdoor scenes....and hybrid scenes. It will save you a ton of time and experimentation.

Is this video training the free one's he has posted on youtube? Or are they something for purchase? Link?

Chris S. (Fez)
03-03-2018, 03:26 PM
The introductory videos are free on youtube. The commercial tutorials I am referring to are not yet available on his site. You can likely contact him directly to purchase an earlybird version. Well worth it.

http://forums.newtek.com/showthread.php

Asticles
03-03-2018, 03:50 PM
You should isolate first which is the source of the fireflies. (not where they are, but the emission that makes them possible) IMHO.

Edit:As Rebelhill explained to me on other scene, try to set Raytrace only on Reflection options, in shader model tab of each material.

Axis3d
03-03-2018, 05:57 PM
You should isolate first which is the source of the fireflies. (not where they are, but the emission that makes them possible) IMHO.

Basically, they are appearing on any surface which has a roughness value higher than zero. I have switched off Sample Backdrop in the Global Illumination Tab and made sure that all my materials are ray trace-only now. And I'm using an environment light which is sampling the backdrop (which is a small, blurred HDR, no high contrast to it).

This render seems to have gotten rid of most of the fireflies and is acceptable enough, but the render time with the environment light is about double (about 37 min.) than my previous setup. I did do another render after this one, with higher camera samples, to clean up some of the grain in the distant building, but that just introduced a few fireflies in different areas - weird. You would think that adding camera samples would just clean up what I already had, not add more fireflies.


140491

Asticles
03-03-2018, 06:19 PM
And they are on the specular indirect pass?

Chris S. (Fez)
03-03-2018, 06:22 PM
If the edge AA was fine and you are just getting rid of grain then up the samples of the lights and such. Adding camera samples is inefficient in 2018 compared to optimizing particular sample parameters.

RebelHill
03-03-2018, 06:36 PM
You would think that adding camera samples would just clean up what I already had]

Adding camera samples only increases rendertime... Ive explained this enough times in enough threads.

Axis3d
03-03-2018, 06:44 PM
If the edge AA was fine and you are just getting rid of grain then up the samples of the lights and such. Adding camera samples is inefficient in 2018 compared to optimizing particular sample parameters.

My camera samples are only at 16, anything less and the edges start to get chunky. And yes, the reflection samples, light samples, volumetric samples, etc. have all been tweaked to get rid of their grain. It seems that the Sample Backdrop in the GI tab was mainly the culprit. Using the Environment light seems to get rid of most of that now. Although, I did like the speed of the 'Sample Backdrop' over the Environment light. But I can definitely deal with this. The quality of the renders with the new PBR is so much better now. I may now have to figure on using a render farm for things now. :)

gar26lw
03-03-2018, 07:43 PM
are there any bugs submitted for this stuff?

Axis3d
03-03-2018, 08:26 PM
are there any bugs submitted for this stuff?

I'm not sure if this is a problem inherent in all PBR software or if this is a question of proper workflow in LW. Does Octane or other PBR's suffer from this?

mummyman
03-12-2018, 09:11 AM
I'm not sure if this is a problem inherent in all PBR software or if this is a question of proper workflow in LW. Does Octane or other PBR's suffer from this?

Any more luck? Just curious, as a week or so ago, I had a glassy figure.. higher roughness and was getting a few annoying fireflies, but in the refraction buffer. Tried many different tuts and sample numbers. Couldn't figure it out yet.

Axis3d
03-12-2018, 12:46 PM
Any more luck? Just curious, as a week or so ago, I had a glassy figure.. higher roughness and was getting a few annoying fireflies, but in the refraction buffer. Tried many different tuts and sample numbers. Couldn't figure it out yet.

I haven't seen anyone else comment on the comparison between different software. I only ever used the trial version of Octane, but never experienced the firefly issue. Only grain, which was never that bad. I managed to get most of the fireflies out of what I was working on by the workflow previously described. But the renders are definitely longer (but look much better than the 2015 renderer). I haven't tried anything really with stuff that has glassy, refractive surfaces, other than calling up some of the LW content scenes. To me, it seems like a bug. But I would like to know if this is the 'final' version of the renderer.

MonroePoteet
03-12-2018, 04:46 PM
This is probably totally shooting in the dark, but for image renderings that have firefly problems, does the location / intensity of the fireflies change if you increas the output image resolution *by ONE pixel*? Or maybe 2-3 pixels? I.e. if you're rendering to 4992x4992 pixels, try 4993 or 4994 or 4995 and see if the fireflies change (NO, I'm not proposing this as a solution, just a debugging tool).

I haven't been able to reproduce the fireflies, but watching the rendering engine migrate its "little squares" across the output image, an off-by-one error in calculating / iterating the "little squares" might cause it to miss hot spots in the specular "reflections" and leave the fireflies behind. Maybe the iteration needs to overlap the regions by a pixel or two to avoid missed spots.

Also, if anyone is having problems with fireflies and using Distant lights, does changing the Distant lights to other, even temporarily, remove the fireflies?

As a said, probably a shot in the dark. Ah well.

mTp

rustythe1
03-12-2018, 04:56 PM
yea, if you use area and environment lights, you absolutely don't get them (or at least i don't), i can set everything to 1 and never get them, when i have had them its been with the distant light only and a scene that contains bad geometry like chair cushions and such, ive only really had them in 2 scenes since using 2018

mummyman
03-13-2018, 06:13 AM
Both sound interesting.. I'll try to investigate this week. My model is maybe an exported figure from Daz or something. No clue where its actually from. It's a human woman figure. Doesn't look horrible, but could be holes or something weird. And I've tried enviro lights and spherical. I'll try swapping out geo too. Interesting.


Changing my camera size from 1280x720 to 1282x722 def changes the fireflies. Not really for the better.. but makes them jump areas. I guess that makes sense

MonroePoteet
03-13-2018, 09:12 AM
Well, to me the LW2018.0.1 renderer's interpretation of Specular seems broken. With *very* small values of Specular on the default Principled BSDF surface, it produces pure-white hot spots for various types of Lights, even if the Light isn't set to "Visible to Camera", GI is off, and the Camera color isn't white. So, I'm guessing that it can produce hot spots on a lot of Surfaces if the normal of the calculated subpatches / micropolygons are at an appropriate angle between any Light and the Camera.

For example, here's a simple Unit sphere (1m in diameter), tilted so one of the panel's normal is appropriate to reflect the Light to the Camera. For a Distant Light, the issue is quite severe, IMO. The first image here has Specular set to 0.0%, then 0.1%. There's an immediate pure-white hot-spot on that panel:

140629 140630

and gets much worse at the default Surface Specularity of 50%:

140639

Even if the color of the Light is changed, the hot spot is white:

140631

With Spherical or Point lights, the behavior is still there, although at slightly higher Specular values:

Point Light, Green color, 10% Specular:
140632

Spherical Light, Green color, 10% Specular:
140633

To me it seems that the new Renderer's interpretation of Specular seems to be scaled wrong (see 0.001% radical change below), and with GI turned off, I don't know where a white hot spot would come from with a Green light, but maybe I'm wrong. For comparison, here are similar images using the LW2015 renderer - yes, I can get a large white hotspot when Specular is 100% and Glossiness is 100%, but it's never as concentrated as the LW2018 renderer:

140640 Specular 0.0%, All are at Glossiness 100%
140641 Specular 1.0%
140642 Specular 10.0%
140643 Specular 100%
140644 Specular 100%, Green Light


Here's another example, using a flat 5x5 segment Box that's been slightly stretched in the Y axis. Note the radical change (IMO) from 0.0% Specular (again, GI is turned off):

140634

to 0.001% (1 one-thousandth percent!):

140635

with very little change to 5.0% or 100.0%:

140636140637

The change from 0.0% to 0.001% is even more radical with GI turned on:

140638


So, I think the new renderer could produce pure-white hotspots on ANY surface if the normal of the subpatch or calculated micro-polygon is appropriate and Specularity is non-zero. One purpose of the migration of the "little squares" (that's a technical term :) ) seems to be intended to eliminate these hot spots (which are visible on any rough surface before they begin), so I'm guessing there's an issue with the overlap between the "little squares" which misses some of the produced hot spots.

To eliminate them, I think LWDG will need to add one pixel (or a couple) to each edge of each "little square" to make sure any of these hotspots are eliminated. Or, fix the (IMO) broken scaling on Specular which produces them in the first place.

Anyway, just a wild guess! I think the white hotspots are not appropriate, especially with a different colored light and how small Specular has to be to get them, but I could be wrong.

mTp

Snosrap
03-13-2018, 11:20 AM
A 1% setting of specular is as bright as 100% specular - only smaller. But change the roughness and you will see significant differences. With a perfectly smooth surface any amount of specular settings will be super bright. It's the microscopic bumps on surfaces that break up those reflections - aka roughness setting. IMO this is all correct. The "stray" fireflies on the other hand don't make sense to me as I've seen them on perfectly muted materials for no apparent reason.

jwiede
03-13-2018, 12:00 PM
A 1% setting of specular is as bright as 100% specular - only smaller.

That doesn't explain the hotspot always being white even with colored lights, nor what happened with the 0.0%->0.001% example Monroe did above.

The specular hotspot color should match whatever light color is used, as bounce off a gray surface. That it's always white is definitely a bug. At such extreme low levels (0.001%) it's also generating too much effect (as Monroe's example above shows), likely why it "maxes out" such that 5%-100% vary so little visually.

MonroePoteet
03-13-2018, 02:38 PM
Here's a LW2018.0.1 rendering of a Blue sphere with a Distant Yellow light at .1lux on a Black background. A bright white hotspot in this instance seems really excessive and broken to me. The Specular Tint is set to 100%, so until the hotspot occurs, the "reflection" is appropriately green.

140659

140658

Here's the same scene with a 3.14lux (default) intensity Yellow light:

140660

RE: increasing the roughness, I'd guess that in a "physics based renderer" that roughness is implemented with "micro-facets" at random orientations depending on the Roughness value. A Roughness of 0.0% means "mirror like", while a Roughness of 100% orients the micro-facets at significant angles to the polygon's normal, scattering the light off the micro-facets.

The basic point here is that the micro-facets are not infinitely small, and if the normal of any micro-facet is at an appropriate angle between the Light and the Camera for this (again, IMO) bizarrely excessive hot-spot to show, it'd appear as a firefly.

I'm reasonably certain there's a bug caused by a boundary (off-by-one) error calculating / iterating the "little squares" (GOTTA be a better term for this - current rendering spot iteration, maybe?) so there's a one-pixel gap between them under some circumstances, causes any fireflies to be missed.

If someone can upload a scene / object that has fireflies (I've still yet to reproduce it with my simple scenes), I could experiment.

As it is, someone having troubles with the fireflies might temporarily set the Specular on ALL of their surfaces to 0.0% and see if they go away.

mTp

BeeVee
03-13-2018, 02:45 PM
Can I ask why you're still using 2018.0.1? If there *is* a bug, it may have been corrected in the 2018.0.2 patch. After all, there were more than 200 bugs and suggestions corrected and implemented in that version.

B

MonroePoteet
03-13-2018, 03:34 PM
Can I ask why you're still using 2018.0.1? If there *is* a bug, it may have been corrected in the 2018.0.2 patch. After all, there were more than 200 bugs and suggestions corrected and implemented in that version.

B

Good point. Sorry! I was out of town when LW2018.0.2 came out with my primary Internet access being a slow Internet Hotspot, so a 257MB download was daunting. Then the download / install slipped through the cracks.

Here's are LW2018.0.2 renders of a Gray ball with a Yellow light. With Specular Tint set to 100%, the white hotspot still seems out-of-place and broken to me, but I could be wrong. As I said, I've yet to reproduce the fireflies themselves with my simple experimenting.

140661 1% Specular, 5% Roughness
140662 10% Specular, 10% Roughness

Reducing the Intensity of the Light does seem to have been addressed, reducing the hot spot to something akin to what might be expected at 0.1lux:

140664 0.1lux
140663 3.14 lux

mTp

BeeVee
03-13-2018, 05:10 PM
other way around for the last two images, but I get what you're saying ;)

B

MonroePoteet
03-13-2018, 05:15 PM
Yep - the upload seems to reverse multiple file uploads, doing the first one in the list last. I usually check and ensure the captions are correct by Preview Post. I'll fix it (if it'll let me) to avoid future confusion!

Thanks,
mTp

p.s. fixed caption order in prior post

rustythe1
03-14-2018, 03:30 AM
so is it always white because you are not setting a specular tint and specular sheen setting (as these are the settings that are similar to the advanced tab old settings for coloured transparency and reflections)

rustythe1
03-14-2018, 03:37 AM
p.s. in your first test you used a distant light, this now has a physical size attached to it in the form of a sun disc, so you would expect quite a large bright spec area, if you make it visible depending on the light placement it can be quite big, is colour space also an issue here (I'm not an expert there but as far as i could tell it can multiply light values if not set up correctly)

rustythe1
03-14-2018, 03:43 AM
p.p.s is the affect the same if you use a 10m sphere instead, i.e the lights are designed to work in a fixed real world scale so does making your objects bigger reduce spec and fireflies?

adk
03-14-2018, 06:45 AM
From my point of view the specular interpretation seems extremely unintuitive. Especially when compared to other engines.
It transitions from 0 to high glancing specularity at a miniscule value of 0.001% (the pop in the GIF), stays virtually the same from there till 2%, then peaks out at what I'd consider a mid glancing and overall specularity, to my eye anyway.
From which point you need to dial up the metallic value to have any hope of making it more specular / reflective.
I want it more specular / reflective, not rougher, which seems impossible to achieve with principled ?
140680

MonroePoteet
03-14-2018, 07:36 AM
so is it always white because you are not setting a specular tint and specular sheen setting (as these are the settings that are similar to the advanced tab old settings for coloured transparency and reflections)

In my tests, Specular Tint was set to 100%. In the earlier tests, Sheen was also set to 100%, but I didn't notice any difference in the white hotspot, so I left it zero in later tests.


p.s. in your first test you used a distant light, this now has a physical size attached to it in the form of a sun disc, so you would expect quite a large bright spec area, if you make it visible depending on the light placement it can be quite big, is colour space also an issue here (I'm not an expert there but as far as i could tell it can multiply light values if not set up correctly)

The "Visible to Camera" flag is clear in all my tests, so I'd expect the "sun disc" to not be rendered, but maybe I'm misinterpretting that setting. Here's a render in LW2018.0.2 with the Distant Light almost 3 kilometers away with Specular Tint and Sheen set to 100%. If there *is* a sun disc associated with the Distant Light, I'd expect it to shrink with distance.

140681

140682

mTp

mummyman
03-14-2018, 08:02 AM
If anyone wants to take a look, feel free.. I've tried double sided, different lights, etc. Still getting fireflies in refraction buffer. I usually chalk it up to me and user error.. but usually it's not this challenging setting up scenes and getting a clean enough render. Something about high roughness and the shape it doesn't like. Maybe the HDRI image. 140683

rustythe1
03-14-2018, 01:41 PM
In my tests, Specular Tint was set to 100%. In the earlier tests, Sheen was also set to 100%, but I didn't notice any difference in the white hotspot, so I left it zero in later tests.



The "Visible to Camera" flag is clear in all my tests, so I'd expect the "sun disc" to not be rendered, but maybe I'm misinterpretting that setting. Here's a render in LW2018.0.2 with the Distant Light almost 3 kilometers away with Specular Tint and Sheen set to 100%. If there *is* a sun disc associated with the Distant Light, I'd expect it to shrink with distance.

140681

140682

mTp

i think lights are still visible in reflections and spec even when not camera, and I'm not sure that the position of the light matters for distance light, its only the rotation as i think the sun stays a set distance from the centre of the LW universe and is based on the environment, see here when you rotate the light the light disc rotates along the horizon, if you move the light, the disc stays in the same place, so if the disc is providing the spec, its always bright white and would appear larger than a point light, if your getting the same with a point light then its probably a bug, but looking at that spec hit in your grab, it is disc shape on a flat area, also the fact it stays white is a problem, but maybe there is another obscure setting somewhere that's the issue
140685

MSherak
03-14-2018, 02:02 PM
If anyone wants to take a look, feel free.. I've tried double sided, different lights, etc. Still getting fireflies in refraction buffer. I usually chalk it up to me and user error.. but usually it's not this challenging setting up scenes and getting a clean enough render. Something about high roughness and the shape it doesn't like. Maybe the HDRI image. 140683

Turn off Glossy Reflections in the shading Model tab. This is what is causing your fireflies. Main problem with 2018 is that the user can't control the reflections when this button is on.


Try these settings.
------------------------
Color 255,255,255
Roughness 45
Specular 0
Transparency 99
Transmittance 8,5,43
Transmittance Distance 50cm
Refraction Index 1.42


One will notice that it's shiny somehow even though specular is off. It's the thing I hate about Principled shading. Everything really is fixed and the user has no real control. (Locked Math)

mummyman
03-14-2018, 02:30 PM
Turn off Glossy Reflections in the shading Model tab. This is what is causing your fireflies. Main problem with 2018 is that the user can't control the reflections when this button is on.


Try these settings.
------------------------
Color 255,255,255
Roughness 45
Specular 0
Transparency 99
Transmittance 8,5,43
Transmittance Distance 50cm
Refraction Index 1.42


One will notice that it's shiny somehow even though specular is off. It's the thing I hate about Principled shading. Everything really is fixed and the user has no real control. (Locked Math)

OH man.. thanks so much. Yup.. that worked wonders. Been driving me batty. Was frustrating the heck out of me. I'll try to remember this one!

djwaterman
03-14-2018, 04:01 PM
in all these tests the sphere hasn't been smoothed, I don't understand why you would do that, how can you make judgments on the fidelity of reflections and such on faceted geometry?

dyls_E
05-13-2018, 03:38 AM
it's the one thing i really can't stand about 2018... i had a job the other day i needed looking good, so i thought stuff it, it's not that big just turn everything up higher than what you think you would possibly need. Took 30 hours to render had hardly any reflection and was still full of fireflys. Took about an hour in photoshop with the heal brush just dotting each one out... considering going back to 2015 till i understand the render settings.

CaptainMarlowe
05-13-2018, 12:33 PM
If you have OD_Tools, there is an image filter called defly fast which does a decent job getting rid of fireflies. It doesn't clean all of them, but is an interesting time saver.

samurai_x
05-14-2018, 04:09 AM
it's the one thing i really can't stand about 2018... i had a job the other day i needed looking good, so i thought stuff it, it's not that big just turn everything up higher than what you think you would possibly need. Took 30 hours to render had hardly any reflection and was still full of fireflys. Took about an hour in photoshop with the heal brush just dotting each one out... considering going back to 2015 till i understand the render settings.

Most of the issue is from reflection samples ime. Turning it up to 8 and up reduces it. But beware of long render times above 12.
Turning off glossy reflectiions sucks. It makes the render ugly. If you can live with that go for it.
There's no other way to speed up the new renderer unless it goes GPU or get AI denoiser.

SaT
05-15-2018, 01:40 PM
Most of the times I get fireflies when I use the distant light at 0 - 2. But I found out that if you increase the angle of the distant light (about 20) you get very good results and the fireflies are gone. Of course the light casts slightly softer shadows but it is still like sunlight and you can see the shadow clearly.
The best about this is that you do not need high reflection samples and no high AA. So it is still very fast.

rdolishny
05-17-2018, 10:53 AM
That is looking good now tho!

rdolishny
05-17-2018, 11:07 AM
Has anybody rendered an animation to see if there is any rhyme or reason with the location or regularity or intensity of these fireflies?

I'm AFK until tonight but will definitely do a test on 2018.4 and report back. I had a rendering with a glass tube with a firefly on my stills. It's time for animation.

If the fireflies move with the object, it's an problem with polygons or surfaces.

If they don't move even though the objects are moving, it's a problem with the camera.

If it's totally random for every frame it's a problem with the renderer globally.

Airwaves
06-26-2018, 10:39 AM
Has any of these fireflies issues been fixed or will they be fixed. It is driving me crazy. I have a simple scene setup with two logos and a rotating stone. I have two sets of lights and in the properties panel of the lights one set of lights is turned off for one logo and the other for the other. (hope that makes sense)

The lights are set up in an envelope to be dark and then come on slowly. The first logo appears and then a few seconds later the next words appear. During this time those stupid lightwave specks show up when the words below are not even on the screen. Can anybody tell me they are trying to fix this or is this the new Lightwave? (I don't mean that in a mean way but need to know if this is the new norm)

https://www.youtube.com/watch?v=_aHsxRN2rWw

They are hard to see but I can see them flickering at the first of the video. I can also see them in the still images from the render as well.

Asticles
06-26-2018, 12:25 PM
To me the best solution to fireflies is to set limit dynamic range to 1

Airwaves
06-26-2018, 01:34 PM
Limiting Dynamic range did fix it. Thank you.

But what I don't get is if the setting is to have all light off to the model why would it show any fireflies?

jwiede
06-26-2018, 03:22 PM
To me the best solution to fireflies is to set limit dynamic range to 1

That's not even a generally-applicable workaround, though, let alone a solution.

Axis3d
09-09-2018, 09:12 AM
To me the best solution to fireflies is to set limit dynamic range to 1

I have tried the Limit Dynamic Range option, only because I could not get the random fireflies out of the render with all of the other options mentioned.

It works very well, except if you are doing multi-pass rendering. It seems to only affect the final rendered image and not the actual render passes (specular direct, specular indirect).

This would be a perfect option if I can figure a way to have it affect the render passes.

jwiede
09-09-2018, 04:45 PM
I have tried the Limit Dynamic Range option, only because I could not get the random fireflies out of the render with all of the other options mentioned.

It works very well, except if you are doing multi-pass rendering. It seems to only affect the final rendered image and not the actual render passes (specular direct, specular indirect).

This would be a perfect option if I can figure a way to have it affect the render passes.

So you believe most have no need for HDR image render output, then?

Tim Parsons
09-09-2018, 06:34 PM
After 8 months of using LW2018 I can happily say that fireflies are mostly a thing of the past. If you are loading older assets look at the bump maps to see if they are causing your fireflies. Also don't be afraid to crank up your keys lights to 32 samples. Some higher sample settings really effect render times, while others not so much.

mav3rick
09-10-2018, 04:20 AM
and than we come to reflection noise...

Axis3d
09-10-2018, 06:45 AM
So you believe most have no need for HDR image render output, then?

I don't know what needs most people have when rendering. I do know that this method worked in this particular case. Also, most people may assume that you can only limit the dynamic range from 0 to 1. You can enter a number higher than one and the images are still HDR. I was able to use this setting as sort of a clamp to eliminate the brightest of the fireflies.

Also, as I've said, it doesn't affect the render buffers and only affects the final output. This is not my preferred method, but in this case it was much better than cranking up sample settings and increasing render times.

Obi
09-23-2018, 10:09 AM
hello,

just about the 'firefly' an example;
(gray color is chrome)

142873

all light; volumetric samples;34 samples; 34

the mac version crashes is a horror, as soon as you change the light settings manually

Axis3d
09-23-2018, 11:19 AM
hello,

just about the 'firefly' an example;
(gray color is chrome)

142873

all light; volumetric samples;34 samples; 34

the mac version crashes is a horror, as soon as you change the light settings manually



Do you need volumetric samples set so high? Are your lights volumetric? That setting is for the graininess seen in the fog from the lights. You could try adjusting your speculator samples.

Obi
09-23-2018, 11:43 AM
Do you need volumetric samples set so high? Are your lights volumetric? That setting is for the graininess seen in the fog from the lights. You could try adjusting your speculator samples.

no volumetric, but I search speculator samples...

Kryslin
09-23-2018, 02:09 PM
I think he means reflection samples, under the render tab on the render properties panel.

Asticles
09-24-2018, 03:14 AM
Hi, is this carpaint shader? It makes a lot of fireflies. If so, change it to principled.

Regards

Obi
09-24-2018, 05:11 AM
Yes Asticles, carpaint shader
Kryslin; Thank you

pnevai
04-29-2019, 02:27 PM
What are your lighting settings? My first guess would be that you are sampling the backdrop from multiple sources which is causing too high contribution to indirect specularity.

If using environment light, don't sample backdrop in GI, same goes if you use portal lights. At the same time, if you are not sampling the backdrop in GI switch the shaders to raytrace reflection and refraction only.

Below is a link to an excellent tutorial on how to get rid of those pesky fireflies in you renders. It goes through step by step on how using environment lights can get rid of those rendering artifacts.

https://www.youtube.com/watch?v=72AAwSFx4nA

pnevai
06-27-2019, 05:02 PM
What he said.... I found it helps dramatically with the firefly issue.... Turn off glossy reflections.