I've been using the Denoise node in the compositor, hooked up to Denoising Normal/Albedo, with those render denoising passes turned on. Works fine. I believe that's the OpenImage denoise algorithm.

In latest 2.91 nightlies, I see in the Scene settings under Denoising, a Render option with choices NLM and OpenImageDenoise.

How are these related? Are they the same? If I keep my denoise node in the compositor and turn on the Scene-settings denoise do I get two denoise passes? If I choose NLM in the settings, what happens to the denoise node in the compositor? Should I just remove the compositor nodes?

I see in this commit the new text explaining that the denoise passes are adapted to the selected denoiser, but that doesn't say anything about what you need in the compositor.

And finally, is OpenImageDenoise or NLM better?


1 Answer 1


The short answer is that if you enable the Denoising in the render options you do not need to use the compositor Denoise node. And OpenImageDenoise is better.

The longer answer is that OpenImageDenoise is the same as the compositor node but it works while rendering rather then after the the render which can speed things up a lot. NLM is a denoiser that doesn't use ai unlike the others so it isn't as good.

This is a version of a project i started that has subsurface scattering so it is harder to render

This is with only 5 samples and no denoising This is with only 5 samples and no denoising

This is all the same and still with 5 samples but with NLM denoiser which is the one that doesn't use ai NLM denoise

And this is identical but with the OpenImageDenoiser made by Intel that uses ai but can sometimes be slightly slower than Optix ai denoiser by Nvidia that is only for compatible Nvidia graphics cards and slightly less impressiveOpenImageDenoise As you see the OpenImageDenoise is better

For curiosity's sake this is the faster but only for Nvidea cards Optix real-time denoiser that is somewhere in between NLM and Intel OpenImageDenoise Optix


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .