Capture sharpening in darktable 5.6 can increase micro-contrast in detailed structures such as hair and wood, but it can also amplify sensor noise and make halo artifacts more noticeable. Treat it as an optional RAW-stage correction, not a guaranteed recovery tool: check white balance, color fringes, and shadow noise first, then inspect the contrast-sensitivity analysis and compare the result with sharpening off. Lowering the iteration count only reduces the overall sharpening strength; it does not selectively exclude noise or correct chromatic aberration. Correct visible color fringes and reassess the mask and on/off comparison first. If grain, color fringes, or halos remain unacceptable, turn capture sharpening off.
Decide and proceed in order
The demosaic module reconstructs full RGB pixels from the camera sensor’s color-filter data. Capture sharpening is available there because some of its processing requires RAW CFA data; the sharpening itself is applied to demosaiced RGB data. The darktable demosaic manual describes it as a way to recover detail softened by sources such as diffraction or an anti-aliasing filter. It is intended to improve micro-contrast in detailed structures, not to act as general sharpening or local-contrast enhancement.
Use this sequence for a RAW image whose hair or wood looks soft:
- Inspect the unsharpened rendering at a useful zoom. Decide whether the texture is present but subdued, or whether it is obscured by motion blur, missed focus, or another problem. Sharpening is not a repair for motion blur and cannot establish detail that the RAW data does not resolve.
- Check that white balance is reasonable. The manual lists decent white-balance parameters as a prerequisite for optimal results.
- Inspect high-contrast boundaries for color fringing. If chromatic aberration is visible, address it with the raw chromatic aberrations module before judging capture sharpening; the manual warns that low chromatic aberration matters and that stronger halo artifacts may still occur.
- Enable capture sharpening. To calculate the radius for the current image from its RAW sensor data, click the button to the right of the radius slider. The manual says this calculation works better with the image fully zoomed out and uncropped in Darkroom, and may fail with substantial chroma noise. Inspect the result; avoid manually increasing the radius far beyond the estimate, where halos can appear.
- Use the button beside the contrast-sensitivity slider to visualize the mask produced by the local-variance, luminance, and threshold analysis. Inspect whether the analysis appears to distinguish the target hair or grain from noisy shadow areas; do not treat the mask as proof of which pixels are sharpened or that the result is artifact-free. The manual notes that very noisy images require larger threshold values than the default, while very low-noise images may benefit from a lower threshold in very dark areas. Judge any adjustment by the mask and an on/off comparison for this image.
- Compare the same crop with capture sharpening on and off. Keep it only if the target texture reads more clearly without objectionable new grain, colored edges, or halos.
This same-crop, on/off comparison is an editorial inspection method, not a documented darktable test procedure. Capture sharpening is distinct from later creative sharpening and output sharpening, and it is not a motion-deblur tool.
A hypothetical acceptance worksheet
The observations below are illustrative, not measurements from darktable or promises about what a setting will do. They show how to turn the mask and image evidence into a decision without prescribing universal values.
| Check | Hypothetical observation | Decision |
|---|---|---|
| Target texture | Individual wood grain lines are visible in the unsharpened view but look subdued | A cautious capture-sharpening comparison is relevant |
| White balance | Neutral-looking wood still has a plausible color cast for the scene | Continue checking; do not treat sharpening as a color correction |
| Color fringe | A colored rim appears along one high-contrast hair edge | Correct RAW chromatic aberration first, then reassess |
| Shadow noise | A dark background shows visible speckling before sharpening | Check whether the analysis suggests reduced participation from noisy shadows, then verify with the on/off comparison |
| Mask and result | The analysis appears to distinguish hair from noisy shadows, and the on/off comparison adds no distracting rim | Retain capture sharpening for further processing, subject to later review |
| Failure branch | Grain, color fringes, or halos become more conspicuous, or the hair looks no clearer | Correct visible color fringes and reassess the mask and on/off comparison. Lowering the iteration count only reduces overall sharpening strength; if artifacts remain unacceptable, switch capture sharpening off. Do not assume a threshold change will exclude a particular noisy area |
The visualized mask shows the result of the contrast-sensitivity analysis; it should not be read as a guaranteed map of the pixels that will be sharpened or as proof that noise is eliminated. The manual describes analysis based on local variance, luminance, and a user-defined threshold. Very noisy images require larger values than the default, while very low sensor noise may allow a lower threshold to improve capture sharpening in very dark areas. These are conditional controls: inspect the mask and compare the rendered result for the image at hand.
Limits and what to try next
If color fringes are visible, correct RAW chromatic aberration first; then assess whether the noise is acceptable and inspect the contrast-sensitivity analysis alongside an on/off comparison. Lowering iterations reduces overall sharpening strength, but does not selectively exclude noise or correct color fringes. If grain, fringes, or halos remain unacceptable, turn capture sharpening off. For motion blur, use another suitable frame where possible; capture sharpening is not a motion-deblur tool.
Demosaicing itself can affect fine-detail rendering and colored edges. The manual notes that demosaic algorithms differ in their handling of high-frequency content, noise, and color overshoots; choosing a different algorithm is a separate decision from enabling capture sharpening. If moiré or maze-like artifacts are present, inspect the demosaic result rather than assuming stronger sharpening will help. Do not use capture sharpening as a substitute for later creative or output sharpening: those have different jobs and should be judged at their respective processing or delivery stages.
Does the radius estimate mean this image is optimally sharpened?
No. It is an automatic estimate based on RAW sensor data, and the manual notes a condition under which it may fail. Check the result and halos rather than treating the estimate as a guarantee.
Does the contrast-sensitivity mask show exactly which pixels are sharpened or remove all noise?
No. The manual says the mask visualizes the contrast-sensitivity analysis, which uses local variance, luminance, and a user-defined threshold to decide which regions are considered safe for sharpening. Treat it as an analysis aid, then inspect the rendered on/off comparison for amplified noise.
Should I increase the radius until hair looks crisp?
Not as a general rule. The manual cautions against raising it much beyond the automatic estimate because halo artifacts can result. If no useful texture emerges, turn the module off rather than chasing crispness.
Sources
This is an original workflow guide based on the linked official documentation. The example worksheet is hypothetical; no software test or image-specific result is claimed.
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.