Iray new Fireflies issue

Just uploaded a two weeks ago to 4.12 Pro. Can't remember which one I had prior but it was the last build before Dforce. I started getting fireflies in my renders a couple days ago. Never had an issue before. I havn't changed any of my render setting which weren't really modified at all except for the size of my render window. I'm using a Gforce 1080 Ti and its on a Windows 10 Intel i7-7700kCPU, 32GB Ram. Here is a recent shot. lots of fireflies around the eyes and the denim shader (which seems really glossy for denim). Its been happening on skin hair and some clothing shaders mostly. I usually use HDRI lighting. No spotlights.

Any suggestions? Thanks

test.jpg
1500 x 1200 - 1M

Comments

  • spent some time doing osme test. My old scenes with G3 characters render find. Its the new G8 charcaters that are getting fireflies. So I went through the surface setting and found that Metal Flake is getting alot of fire flies but evenwhen that's turned off, I still get some random fireflies on her. 

    Sure would appreciate anyone helping me figure out what's happening with the new G8 models' shaders and getting back to good renders. I really like what I see in the G8 stuff, but I gotta get rid of these white dots! thanks to any folks that can lend a hand.

  • fastbike1fastbike1 Posts: 4,078

    @sergeantr_c12737cf4c

    Your attachment, even at full resolution isn't showing any fireflys to me.

  • Sven DullahSven Dullah Posts: 7,621
    edited December 2019
    fastbike1 said:

    @sergeantr_c12737cf4c

    Your attachment, even at full resolution isn't showing any fireflys to me.

    So would you agree they look like tiny white dots?

    Post edited by Sven Dullah on
  • Sven DullahSven Dullah Posts: 7,621

    spent some time doing osme test. My old scenes with G3 characters render find. Its the new G8 charcaters that are getting fireflies. So I went through the surface setting and found that Metal Flake is getting alot of fire flies but evenwhen that's turned off, I still get some random fireflies on her. 

    Sure would appreciate anyone helping me figure out what's happening with the new G8 models' shaders and getting back to good renders. I really like what I see in the G8 stuff, but I gotta get rid of these white dots! thanks to any folks that can lend a hand.

    Sorry I'm no IRay user so can't be of much help...I'm sure someone has a better idea of what's going on here.

  • fastbike1 said:

    @sergeantr_c12737cf4c

    Your attachment, even at full resolution isn't showing any fireflys to me.

    Thanks for taking a look. They are small but ther're there. Never had this before upgrading and going with G8. I zoomed in and attached.

    test2.jpg
    285 x 358 - 113K
  • fastbike1 said:

    @sergeantr_c12737cf4c

    Your attachment, even at full resolution isn't showing any fireflys to me.

    So would you agree they look like tiny white dots?

    I attached a zoomed version and circled the dots. See my response to Fastbike1. Thanks fo taking a moment to post.

  • I may have solved my FireFly issue with 4.12. I've been experiemnting with all the render settings. I have Firefly filter on and it was ticking me off to have all these fireflies (had even more in some of my architectural stuff). I turned on Post Denoiser and some of the other filters and bam! No Fireflies. Were these things new in 4.12? I don't remember them in 4.9 (I think that's what I had last). If they are new, shouldn't they just have automatically been on? That would have saved me alot of time and head scrating wondering what I was doing wrong and why my renders weren't going so well after upgrade. Image of the filters that helped are attached. 

    test3.jpg
    554 x 481 - 151K
  • JD_MortalJD_Mortal Posts: 760
    edited December 2019

    Not always metal flake that does this. Simple noise in a bump-map or normal-map can also lead to these sub-surface sub-pixel super-bright pixels showing-up randomly. Remember, to get one pixel "correct", IRAY shoots hundreds of beams of light into each pixels sub-pixel and sub-surface components. Within that simple pixel may be hundreds of texture pixel colors and normal-angles. One may hit that bright red pixel and perfect reflection angle, showing the wrong color-average... Because there is no other data to average it out with, yet. It needs a second, third, fourth, fifth... to get a more correct color average of all the various colors hidden within that pixel, by the sub-pixel and sub-surface data hidden below.

    There are "Fireflies", and there is also just sub-surface, sub-pixel noise. You are seeing more sub-surface noise. Turn the filter off, and you will see that it is doing a great job of removing fireflies already. A single photon just happened to hit a reddish, or bright colored sub-surface sub-pixel, and that is the only color it reflected, for that pixel. When another beam hits that pixel, surely on another sub-surface sub-pixel, ti will average those two color values out, killing the high-contrast noisy color.

    A "firefly", is another beast. When the math fails, you can get a locked "0" or a locked "255", (full off, full on), for specific colors. This can be a color that is not even there, but because the math made a calculation error and locked values, you get a nearly solid color, or solid shade, which may take longer to level-out, on average. (When more photons hit that pixel-area, and get correct colors to adjust the wrong one, which never should have been there. Ultimately, it should disappear. However, sometimes they just don't. Thus, the "firefly" setting. A setting which looks for those high-contrast potential errors, and just throws those values away, instead of using them in the actual render. However, that can remove true high-contrast components of a render, like reflections and metallic flake, etc.)

    Using the denoiser can help, because it is essentially "guessing" at what each pixels shade and color is, by pulling values from adjacent pixels that are "known", until the actual pixel at that spot is known better. (EG, if all surrounding pixels are dark green, chances are, that this unrendered pixel is also dark green, unless it gets rendered, and turns out to be bright red... Then, in time, that will ultimately render-out to the correct color, instead of the "guessed", denoised color/shade.)

    P.S. Doing an early termination render with the denoiser... Followed by one without the denoiser, but run to a longer time... You can use the two images, as layers, in a paint program, and use the denoised image to help denoise your other image, indirectly. Use the denoised image as the shade-levels. You can erase areas that you don't want to use as the shade levels. Though, ultimately, if you render it out long enough, you will be back to square one, as the light values from the real render will replace the guesses from the denoiser. Potentially bringing back more noise into the rendering.

    P.S. I think that Optix may also have been "helping" with the issue, but that is now forced off, I believe. So that may also be part of the issue, as to why we see more noise than before, where optix was essentially ON, by default. As well as tweaks done to the IRAY engine, since the denoiser was created. (Maybe they didn't have to do such an expensive rendering pass, which would have resulted in less noise, but consumed a lot more time. Something that would also encourage the use of the denoiser. Being a win-win for IRAY.)

    Post edited by JD_Mortal on
  • sarge74sarge74 Posts: 121

    Thanks for the info JD. Much appreciated.

  • Nice metallic shiny eye make-up. Geez.

Sign In or Register to comment.