Denoise
Remove noise from images on the GPU with median, bilateral, non-local means, FFT gate, or wavelet denoising.
What it does
Denoise softens noise and grain while trying to hold edges. Content In produces Content Out at the same size and type. Strength sets how aggressive the smoothing is: at 0 the image passes through unchanged, and raising it blends in more neighbouring pixels.
Method sets the character of the reduction: Median, Bilateral, Non-Local Means, FFT Gate, or Wavelet, each suited to a different kind of noise such as impulse speckle, general grain, or fine texture. Radius and Threshold shape the neighbourhood size and how much detail is protected from smoothing.
When to use it
Cleaning up a noisy or grainy image before compositing or grading.
Rendered or scanned images carrying speckle noise that must go without losing edges.
A single noise-reduction pass instead of a hand-built blur and threshold chain.
| Parameter | Type | Default | What it does |
|---|---|---|---|
content_in | Raster | — | — |
method | String | "Bilateral" | Denoising method: Median for impulse noise, Bilateral for edge-preserving, NLM for texture detail, FFT Gate for spectral, Wavelet for detail-aware. One of: Median, Bilateral, Non-Local Means, FFT Gate, Wavelet. |
strength | Number | 0.500 | General denoising strength (maps to method-specific parameters) |
radius | Number | 3 | Neighborhood size: Median kernel, Bilateral spatial sigma, NLM search radius |
threshold | Number | 10 | Edge/detail threshold: Bilateral range sigma, FFT gate dB, Wavelet coefficient threshold |
spatial_strength | Field | — | Spatial field controlling where the effect applies (0=bypass, 1=full) |
Gotchas
Edge protection has a limit. If a subject's edges look too soft after denoising, lower Strength rather than compensating with the other settings.
Strength is the control that matters most at present. Start at 0 and ease it up until the noise settles. Pushed too far, fine texture softens along with the noise.
Worked example
Add a Denoise node and wire a noisy image into Content In.
Pick a Method that matches the noise. Bilateral is a solid general-purpose default.
Raise Strength gradually until the noise clears without smearing detail.
Feed Content Out into the next filter or composite step.
See also
Blur · Sharpen · Convolve · Rasters (images)