The practical answer is: start with local laplacian when you need separate control over shadows, highlights, and the middle tonal range. Evaluate bilateral grid when the job is mainly to choose the coarseness of the detail being adjusted, then set contrast and detail. Neither mode can recover texture that is absent from a completely black source area.
Specific judgment and steps
The darktable 5.6 local contrast documentation defines both modes as operating on the Lab L channel. That is a functional description, not a promise that every image will respond visibly in the same way.
- Choose one crop containing the decision-critical content: a textured surface, a bright region, a shadow, and, if possible, a smooth gradient. Keep the crop and viewing zoom unchanged while comparing modes.
- Select local laplacian if the target is tonal separation. Adjust
detailfirst, then inspecthighlights,shadows, andmid-tone rangeone at a time. Use a lowerhighlightsvalue when the bright end needs compression; a lowershadowsvalue can lift shadows locally. - Select bilateral grid if the target is scale. Adjust
coarsenessto decide whether the operation concerns finer or broader detail, then comparecontrastanddetail. Judge which physical structures change, rather than choosing the mode from the control names alone. - Toggle each candidate against the unchanged crop. Record the affected object, whether the bright area is compressed or merely darkened, and whether a smooth gradient develops visible bands.
- Prefer the simpler mode that meets the stated target. If both seem useful, compare them as separate candidates rather than assuming their settings are interchangeable.
Local laplacian is the better first candidate for a scene where the creative brief says “separate shadows and highlights.” Bilateral grid is the better investigation when it says “work at this detail scale.” These are editorial selection rules derived from the documented controls, not additional darktable features.
An acceptance worksheet
Use the following as a hypothetical observation record. The values are example settings, not presets or guaranteed outcomes.
| Target crop | Mode and example controls | Hypothetical observation | Decision |
|---|---|---|---|
| Bright metal beside mid-tone paint | Local laplacian: inspect highlights and mid-tone range separately | Check whether the highlights are pulled down and whether paint separation remains visible | Keep only if the crop confirms the tonal target without unwanted changes |
| Fine woven surface | Bilateral grid: adjust coarseness, then compare contrast and detail; record the values shown in your interface | Hypothetical observation: check whether the weave changes more than the surrounding broad shading | Keep only if the crop confirms the scale-specific target |
| Smooth wall gradient | Local laplacian: mid-tone range extreme | A visible band appears in the otherwise smooth transition | Reject that setting; reduce the range or compare another candidate |
| Near-black object against a bright background | Local laplacian: lower shadows; record the value shown in your interface | Hypothetical observation: check whether the dark object against the bright background is lifted while a uniformly black interior remains without visible texture | Treat any visible change as local lifting, not recovered source detail |
The table records observations to verify in your image; it does not describe guaranteed algorithmic output. For the documented control meanings, see the same official module reference.
Limits and next step after failure
If the first mode creates halos, reversals, or an unnatural edge transition, reduce the strength and inspect the crop at both working and final display size. The documentation describes local laplacian as designed to resist unwanted halos and gradient reversals, but that does not make every setting artifact-free.
If an extreme mid-tone range produces banding, treat the banding as a failed acceptance result. Reduce that control and reassess the tonal target; do not hide the artifact by judging only a thumbnail. If a shadow lift reveals separation only around a dark object against a brighter area, that is consistent with local manipulation. It does not prove that clipped or completely black interior pixels contain recoverable texture. Obtain a better source or accept the information boundary when accuracy matters.
Three short questions
Should I always use local laplacian first? No. Use it first when the requested change names tonal zones. Start with bilateral grid when the requested change names a detail scale.
Does higher detail restore texture? No. The control can increase local contrast, but the documentation does not establish recovery of information missing from the source.
Is a wider mid-tone range safer? No. A wider range classifies more values as mid-tones, and extreme values can produce banding. Judge it on a smooth gradient as well as on textured content.
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.