A slight, static lens defocus may be worth testing with darktable 5.6’s diffuse or sharpen module, using a lens deblur preset as a starting point. It is not a way to reverse motion blur, and it cannot guarantee recovery of detail the image did not record. Before judging the result, correct chromatic aberration and noise, check for clipped blacks, and compare both the intended subject edge and any bokeh that should remain soft.
Decide whether the blur is a candidate
- Inspect the original at 100% and compare several edges. Broad softness without a consistent directional trail can be consistent with defocus; streaked or doubled edges suggest motion may be involved. Treat these as diagnostic clues, not proof. If motion blur is apparent, do not use this module as a motion-deblur remedy: the official module documentation says motion blur is not reversible by undoing diffusion.
- Before deblurring, correct visible chromatic aberration and noise. Check that blacks are not clipped; the manual points to black level correction in the exposure module if correction is needed. It also advises applying diffuse or sharpen after color calibration and before nonlinear operations.
- Apply a lens deblur preset as a starting point. Use 1:1 to inspect local artifacts and how the controls affect a clearly defined subject edge and a separate, intentionally soft background edge; the manual says the preview is exact only at 1:1. Do not treat a sharper-looking pixel-level edge as proof of recovered detail. Make the acceptance decision at the intended viewing or export size, considering whether the subject and the bokeh still look right together.
- Change settings only in response to a visible failure. If the effect is too strong, reduce iterations. If edges develop fringes or halos, increase edge sensitivity. If valid bokeh begins to sharpen, reduce the radius. If deblurring seems correct in bright areas but excessive in dark areas, increase the edge threshold. If deblurring clips black pixels, lower exposure's black level correction as the manual specifies, then inspect the result again; stop if clipping remains or the black-level change is unacceptable. Preset behavior is a starting point, not a promise that every image will respond well.
Use a small acceptance worksheet
The following is a hypothetical observation record, not a report of software testing or a predicted module result. Its values illustrate how to make a decision, not universal thresholds.
| Area or check | Hypothetical observation after the preset | Decision |
|---|---|---|
| Stationary subject edge | A soft edge looks more distinct, with no obvious new fringe at 1:1 | Keep comparing; do not infer recovered texture from edge contrast alone |
| Background bokeh | A soft circular highlight gains a defined rim | Reduce radius; if the unwanted change remains, reject this deblur result |
| Dark region | A shadow boundary acquires a bright or dark edge | Inspect edge sensitivity and threshold; reject if the artifact persists |
| Black pixels | A previously dark area looks clipped after processing | Lower exposure's black level correction as the manual directs, then recheck; stop if clipping remains or the change is unacceptable |
| Intended output size | The subject looks acceptable, but the bokeh no longer matches the intended depth | Prefer the less processed version, even if the subject edge looks sharper |
Record the observation and the decision together. That makes it easier to distinguish a useful improvement on the subject from a global increase in edge contrast.
Know when to stop and what to do next
The module models diffusion with a multiscale process; its radius span selects the range of detail scales it acts on. For deblurring, the manual describes a zero central radius and a radius span approximately matching the lens blur width. This is a starting relationship, not a measurement recipe that guarantees a correct setting. Higher iteration counts may improve reconstruction when speeds are appropriate, but can also make the process slower or cause it to degenerate. If a preset is already too strong, reduce iterations before attempting stronger processing.
Stop when the algorithm starts acting on intentional bokeh, produces persistent edge artifacts, clips blacks, or makes noise more prominent. More sharpening is not evidence of more recovered information. Keep the original as the reference; if the missing detail matters, seek a sharper source or treat any later AI edit as an interpretation, not proof of the original detail. The manual describes the module’s behavior and controls, but does not promise restoration of unrecorded texture or artifact-free results for a particular photograph.
Three quick questions
Is diffuse or sharpen just ordinary sharpening? No. The module can reverse certain diffusion processes, including static lens blur, using the lens deblur presets. Its separate sharpness control is not advised as a stand-alone sharpening method because it does not prevent halos or fringes.
What if the background is supposed to be blurred? Keep it as an acceptance region, not as a defect. If deblurring starts changing valid bokeh, reduce radius; reject the result if the softness cannot be preserved acceptably.
Should I keep increasing iterations until it looks sharp? No. The manual warns that more iterations cost time and can cause degeneration. Stop when artifacts, noise, clipped blacks, or damage to intentional blur outweigh the subject-edge improvement.
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.