Render passes have fireflies, beauty has none
Moderators: ChrisHekman, aoktar
Multipass: diffuse and light ID have many fireflies
In Octane Setting > Camera Imager > hotpixel removal at .82 (no fireflies at this hotpixel setting)
Enable the scene camera's imager, put the hotpixel removal setting all the way to 0, I get less fireflies in multipass but still not not a good image.
What is the best way to deal with hotpixels in a multipass workflow?
- Attachments
-
- FIREFLIES_MULTIPASS_PROBLEM.zip
- C4D file
- (67.55 KiB) Downloaded 123 times
Hi vmedium,
hotpixel removal is a post effect, so it is not good with multipass.
In your scene, the fireflies are generated by the recessed light you have built. This is always the worst scenario for Path Tracing rendering, please, have a look at this other discussion:
viewtopic.php?f=23&t=52982&p=266457&hil ... er#p266457
Here is your scene with your light replaced by a simple plane with IES file in distribution: As you can see, in diffuse pass there are no fireflies, and the Hotpixel removal can be set back to 1.
Note that in the light passes there are still some fireflies, since they are intended to be mixed with other passes, and the fireflies will not be visible.
Anyway, if you need a clean pass, you can render at double resolution, but with a quarter of max sampling for the same time, and downscale at the original size, to kill any one-pixel firefly.
Here is the scene file: ciao beppe
hotpixel removal is a post effect, so it is not good with multipass.
In your scene, the fireflies are generated by the recessed light you have built. This is always the worst scenario for Path Tracing rendering, please, have a look at this other discussion:
viewtopic.php?f=23&t=52982&p=266457&hil ... er#p266457
Here is your scene with your light replaced by a simple plane with IES file in distribution: As you can see, in diffuse pass there are no fireflies, and the Hotpixel removal can be set back to 1.
Note that in the light passes there are still some fireflies, since they are intended to be mixed with other passes, and the fireflies will not be visible.
Anyway, if you need a clean pass, you can render at double resolution, but with a quarter of max sampling for the same time, and downscale at the original size, to kill any one-pixel firefly.
Here is the scene file: ciao beppe
Thanks for the scene. The problem is that GI clamp is applied only to the light passes, which will be fixed with the upcoming 3.08 test 5 release. The diffuse pass should actually be empty and the scattering of the environment medium should be in the SSS pass. And I don't know yet why GI clamp isn't applied to it either, but we will try to fix it as soon as possible.
In theory there is no difference between theory and practice. In practice there is. - Yogi Berra
Thanks to the both of you for answers and updates.
Before you replied, I found a solution (brograph slack)
I switched to DL / GI_DIFFUSE at 3000 samples
It ended up cutting render times in half
Used the camera imager, cranking the hotpixel on camera tag all the way to 0
This gave me firefly free lighting and diffuse passes
RE: Bepeg: I removed the cylinder in my testing altogether, and the fireflies still persisted. Will try your IES. Thanks for the tip on the Double Res with quarter of max sampling. Will try that in the future.
RE: Abstrax... GI clamp not being applied makes so much more sense. I was also confused by the diffuse having the scatter. Thanks for the additional info.
Before you replied, I found a solution (brograph slack)
I switched to DL / GI_DIFFUSE at 3000 samples
It ended up cutting render times in half
Used the camera imager, cranking the hotpixel on camera tag all the way to 0
This gave me firefly free lighting and diffuse passes
RE: Bepeg: I removed the cylinder in my testing altogether, and the fireflies still persisted. Will try your IES. Thanks for the tip on the Double Res with quarter of max sampling. Will try that in the future.
RE: Abstrax... GI clamp not being applied makes so much more sense. I was also confused by the diffuse having the scatter. Thanks for the additional info.