The Spectral AI Denoiser uses machine learning technology to denoise OctaneRender’s output. Its tiling and multi-GPUThe GPU is responsible for displaying graphical elements on a computer display. The GPU plays a key role in the Octane rendering process as the CUDA cores are utilized during the rendering process. support allow denoising of any resolution up to OctaneRender’s maximum, while consuming about 450 MB per device.
The denoiser is also trained to denoise volumes and volume passes. Volumetric passes have very low frequency details, so don't use the Volumetric AI Denoiser with less than 1000 samples if you want to preserve details for final render quality that would resemble a 2K to 10K sample render of the scene.
NOTE: The denoiser cannot be used with the PMC kernel.
In Modo, the Spectral AI Denoiser settings are properties of the Camera item, therefore each camera has its own unique Denoiser settings. The Spectral AI Denoiser settings can be accessed from the Render Toolbar > Camera Button > Imager tab.
Enable Denoising - Enables the Spectral AI Denoiser, which denoises the main beauty and denoiser passes.
NOTE: When enabled, the Denoised Beauty pass will appear in the Render Toolbar > Select Viewport Render Pass popup.
Denoise Volumes - If enabled, the Spectral AI Denoiser also denoises volumes.
Denoise On Completion - If enabled, beauty passes are denoised once at the end of a render. Disable this option if you're rendering with an interactive region.
Min Denoiser Samples - The minimum number of samples per pixel until the denoiser kicks in. This is only valid when Denoise On Completion is disabled.
Max Denoiser Interval - Maximum interval between denoiser runs in seconds. This is only valid when Denoise On Completion is disabled. The Denoiser Interval tells the denoiser to run when OctaneRender® reaches this value.
Blend - Blends the original image into the denoiser output. 0 results in a fully-denoised image, and 1 results in an unaltered image. Intermediate values produce a blend of the denoised image and the original image.