Left: without denoising. Right using imager_denoiser_optix.


The OptiX™ denoiser (based on Nvidia AI technology) is available as a post-processing effect. This imager also exposes additional controls for clamping and blending the result. It is integrated into Arnold for use with IPR as an imager (so that you get a very quickly denoised image as you're moving the camera and making other adjustments). 

There are some requirements to be aware of when using the OptiX denoiser, however:

  1. Ensure that you are using the most recent Nvidia GPU driver version (see here). Note that the OptiX denoiser is not available on OSX.
  2. By default, Arnold will not use the GPU if it has less than 512Mb available when Arnold starts. You can either try to free up GPU resources by quitting applications using the GPU or lower the Min. Memory (MB) settings. The default settings act on the side of caution and won't use your GPU if you have less than 512Mb available so that all GPU resources aren't used or worse try to use more memory than is available on the GPU. Users can change this parameter at their own risk.
  • Arnold will automatically force a box_filter on outputs with filters with known issues when either of the denoisers are used. This is because the OptiX denoiser does not perform well with filters that span across multiple pixels.
  • Denoisers won't work as well after imagers have been applied to them. Therefore denoisers should be applied before any other post-processing imagers (top of imagers list).
  • The OptiX denoiser only works on full frames rather than buckets.

  • The OptiX™ denoiser supports denoising AOVs in a way that allows the denoised result to be compositable.


Input points to the previous imager operation in the chain (so that you can stack them and apply multiple operations serially).


Enables the imager.

Layer Selection 

Selects layers processed by this imager with a glob or operator expression. An imager can use a selection to determine what nodes will be selected to be processed by the imager at render time. The selection is an expression that can consist of multiple selections concatenated by: 

  • or (union)
  • and (intersection)
  • not (negation)
  • and not (exclusion)
  • () for nested scoping


Each selection uses glob patterns and regular expressions and a node will be processed by the imager if the expression matches the node name. By default, glob matching is used unless the selection is in a regex quote, i.e. r'<my_regex>'. For example:

specular or diffuse
not r'sss_(direct|indirect)'
r'color_(mask1|mask2)' or r'mask[34]'

Output Suffix 

When set, creates a new additional output with the input name plus the suffix in which the denoised result will be written into. For example, when denoising the "RGBA" AOV with the output_suffix set to "_denoised", the denoised result will be written into a new "RGBA_denoised" output. It only works with multi-layer drivers.


Linear interpolation between the input and the denoised output.


Clamps the input to the denoiser to a given range.

Clamp Min

Minimum value to clamp to.

Clamp Max

Maximum value to clamp to.

Use Feature AOVs

Enables the use of additional N and albedo AOVs to improve denoising quality. In cases where something other than the beauty is denoised, it may prove better to disable this option.

  • No labels
Privacy settings / Do not sell my personal information / Privacy/Cookies