The short answer is: use raw denoise when you need to address noise in the sensor’s RAW/CFA data before demosaicing; use denoise (profiled) when the noise is being judged in the reconstructed linear RGB image and a usable camera/ISO profile can guide the correction. They are not interchangeable versions of the same filter. Compare them on separate RAW copies, keep the inspected detail and exposure treatment identical, and choose one primary route rather than assuming that stacking both is automatically better.
Specific Decision And Steps
- Identify the data stage.
raw denoiseaccepts linear, scene-referred RAW data and runs before demosaicing. The raw denoise documentation describes itsnoise thresholdand coarse/fine curves: raising the threshold removes more noise but can remove more detail; the left and right ends of the curve act on visibly coarser and finer-grained noise, respectively. Use those ends to adjust the amount of smoothing for the visible noise; do not treat them as a diagnosis of the noise’s physical source.denoise (profiled)accepts linear, scene-referred RGB data after demosaicing. Its module documentation explains that it can use camera-sensor and ISO noise statistics. - Make two comparable versions of the same RAW. Keep white balance, exposure, demosaic method, crop and viewing scale fixed. In the first version, inspect
raw denoisewith a restrained threshold, then move the coarse/fine curve only where the visible noise calls for it. In the second, check theprofilefield indenoise (profiled). darktable can select a camera and ISO profile from Exif data, interpolate between nearby ISO datasets, or allow a manual override. - Classify the remaining problem rather than judging only overall cleanliness. Speckled color, uneven channel response and shadow color casts are different observations from loss of fine edge detail. In profiled denoise, the wavelet
Y0U0V0mode can separate luminance and chroma control;strength,preserve shadowsandbias correctionaddress different trade-offs. Do not infer the noisy channel from a pixel’s visible color alone. - Select the route by comparing both versions on the same critical texture: hair, foliage, fabric, stars or small lettering. Profile availability is one factor, not evidence that raw denoise will work better. If no matching profile is available, profiled denoise can still serve as an RGB-stage comparison, but its configuration basis is limited; if a profile is available, check that it is appropriate rather than assuming it guarantees a good result. Record the visible noise type and detail loss in each version, then choose the route that best preserves the target texture at the intended output size. If either route damages key detail before the noise is acceptable, stop increasing its effect.
Acceptance Worksheet
The following is an author-designed comparison method, not a promised darktable result. The numbers are hypothetical observations for recording a decision, not universal settings or measured benchmarks.
| Checkpoint | raw denoise copy | denoise (profiled) copy | Decision signal |
|---|---|---|---|
| Metadata | Same hypothetical RAW, ISO 6400; a matching camera/ISO profile is available | Same hypothetical RAW, ISO 6400; the same profile is available | Both versions have the same profile basis; compare the denoising results, not profile availability |
| Same detail | 120% view of a dark sleeve with fine threads | Same crop, zoom and exposure | Reject any comparison made at a different scale |
| Noise observation | Fine colored pixels remain near threads | Chroma is lower, but shadow cast needs review | Record noise type, not just “cleaner” |
| Detail observation | Raising threshold softens thread gaps | Lowering strength retains gaps but leaves luma grain | Stop when the required texture is no longer credible |
| Final choice | Record whether the RAW-stage version preserves the target texture and why | Record whether the RGB-stage version preserves the target texture and why | Choose one primary route and record the reason; do not infer superiority from profile availability alone |
Limits And Next Steps
A profile is evidence of supported camera/ISO statistics, not a guarantee that every exposure, white-balance choice or underexposed shadow will look ideal. An intermediate ISO may be interpolated, but that does not remove the need for visual inspection. The manual also describes different algorithm choices inside profiled denoise: wavelets are less resource-intensive, while non-local means can be costly and may smooth fine edges when its patch or search settings are increased.
If no profile is available, first compare a conservative raw-denoise version with the unprocessed noise. If the result still fails, create a separate comparison version in which raw denoise is disabled and denoise (profiled) is the only evaluated primary denoiser; do not leave the other denoiser active in either primary version. A version using both modules is a separate stacking test, not part of this two-route comparison. If both routes erase the texture that matters, keep more noise, revisit exposure and demosaic decisions, or accept that the file does not contain enough recoverable information for the intended output. The demosaic documentation notes that high-ISO images can favor different demosaic behavior, so a denoise failure should not automatically be blamed on the denoise control alone.
Three Short Questions
Should I always run both modules?
No. The modules operate on different data stages, and using both can make it harder to identify which step removed the texture. Test them as separate primary routes first.
Does a visible red or blue pixel prove that channel is noisy?
No. The documentation warns that a visible color can result from noise in other channels. Treat color as a clue and inspect the relevant channel behavior with an appropriate diagnostic view.
When should I stop adjusting strength?
Stop when the protected detail becomes less believable, when color correction starts looking like a cast removal rather than noise treatment, or when additional smoothing only produces a cleaner preview without improving the intended output size.
Sources and next step
- darktable 5.6 user manual - raw denoise
- darktable 5.6 user manual - denoise (profiled)
- darktable 5.6 user manual - demosaic
This guide interprets public documentation and uses explicitly hypothetical acceptance examples, not application tests. Controls and availability can change; verify the cited manual against your installed version.
For work created in Imagild, carry the input requirements and acceptance checklist into the photo editor. The external software controls discussed above belong to their respective products.