After lifting RAW shadows, first confirm that darktable’s denoise (profiled) module selected the right camera profile and ISO. Then classify what became visible: brightness variation, colored blotches, or a green-purple cast in the darkest areas. In auto mode, use “adjust autoset parameters” as a starting point for exposure-related extra noise; in wavelets manual mode, address luminance and chroma separately with Y0 and U0V0. Treat each adjustment as a trade-off with fine texture, not as recovery of detail that the RAW file never recorded.
Diagnose and adjust in order
- Check the profile before tuning. The module normally reads camera model and ISO from RAW metadata and finds a matching profile. If the ISO lies between profiled values, darktable interpolates between the nearest datasets. Inspect the selected profile entry; if the camera or selection is wrong, correct it before judging the denoising response. Profile matching models sensor noise; it does not alter the ISO at which the photograph was captured.
- Compare the same shadow region before and after your tonal lift. Keep the crop and viewing scale consistent. This is a practical comparison method, not a special darktable measurement. Ask whether the change is mostly fine or coarse light-dark grain, colored patches, or a hue shift confined to deep shadows. A pixel’s apparent color alone is not enough to identify a noisy sensor channel.
- For auto mode, increase “adjust autoset parameters” only as a controlled starting point. The control changes the denoising parameters as if using an effective ISO: actual capture ISO multiplied by this control’s value. It does not change exposure, metadata, or the sensor’s recorded ISO. Reassess after each change; do not treat an exposure lift as proof that one particular compensation is correct.
- If the main issue is separable, switch to wavelets and Y0U0V0. Raise or lower the Y0 curve according to unwanted brightness noise, and adjust U0V0 according to colored noise. Wavelet curves act by noise coarseness: the left side affects coarser grain and the right side finer grain. Use the image’s visible structure to choose a direction, not a universal curve shape. See the official wavelet and Y0U0V0 guidance.
- In advanced mode, address shadow-specific symptoms. Lower “preserve shadows” when shadow noise needs stronger treatment; change “bias correction” only if the deepest shadows show a color cast. The documented direction is to increase it for shadows that look too green and decrease it for shadows that look too purple. Retest fine edges and texture after either change.
The manual recommends wavelets Y0U0V0 as the preferred wavelet approach and says the module can handle luminance and chroma separately without two instances. Avoid adding a second instance with a color blend mode as a routine workaround: the module’s pixel-pipeline position makes that approach inappropriate according to the module documentation.
Shadow diagnosis worksheet
The entries below are a hypothetical example of a photographer’s notes, not measured results or recommended numeric settings. “ISO 640” is the capture metadata; the effective ISO entry describes the auto control’s model, not a changed capture.
| Check | Hypothetical observation | Decision to record |
|---|---|---|
| Capture and profile | RAW metadata says ISO 640; the displayed profile names the expected camera | Continue only if the camera/profile selection is credible; note any mismatch for correction |
| Lifted region | A lifted wall shows stronger grain, while edges and texture remain visible | Try a restrained auto compensation or inspect Y0; accept only if the texture remains useful |
| Noise type | Dark flat area has colored blotches as well as brightness grain | Compare U0V0 and Y0 separately; do not use one curve as a diagnosis for both |
| Shadow hue | The deepest corner looks slightly green, unlike nearby midtones | Check bias correction in advanced mode; record the direction tried and whether the cast remains |
| Detail check | Fine wall marks soften before larger blotches disappear | Back off strength or shadow treatment; keep residual noise if further smoothing erases relevant detail |
After each change, compare the same lifted region and a nearby transition containing real detail. A cleaner patch is not an improvement if it also removes the evidence you intended to retain.
Limits and what to try next
The profile is only as suitable as its camera match and available sensor data. The manual describes profiles for many cameras, but it does not guarantee a profile for every model or a perfect match for every file. It also does not define one correct compensation value or curve for all cameras and lifts. A cast, blotch, or grain pattern is a clue to inspect controls, not proof of its cause.
If the selected profile is doubtful, verify the camera identification and profile choice before making stronger adjustments. If wavelet separation does not address the visible structure, compare the module’s other algorithm, non-local means, while considering that it is more resource-intensive. If denoising erases important fine detail before the noise is acceptable, keep a less-smoothed result or revisit the tonal lift; denoising cannot establish detail absent from the recorded RAW data. These are diagnosis choices, not results guaranteed by the manual.
Common questions
Does effective ISO change the RAW exposure? No. In auto mode it is a denoising calculation based on capture ISO multiplied by the adjustment control, not a change to exposure or recorded metadata.
Should I use Y0 for every visible colored speck? No. Y0 controls luminance denoising and U0V0 controls chroma in wavelets Y0U0V0. Decide from the noise pattern across the shadow area, not the color of a single pixel.
When should I change bias correction? When the darkest shadows show a green or purple cast. Follow the documented direction, then check the same shadow and adjacent real detail; do not use it as a general color-balance control.
Sources and next step
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.