
RawTherapee dark-frame subtraction addresses thermal, dark-current and fixed-pattern effects. It is not a general replacement for high-ISO noise reduction. Use a reference RAW from the same camera model and suitable capture conditions, then inspect which frame or averaged template the application selects.
The official Dark-Frame documentation explains the reference directory and automatic matching. A very dark-looking photo from another camera is not a valid substitute.
Identify the defect before choosing the reference
A repeatable bright pixel or fixed pattern in a long exposure is a different problem from random grain across a high-ISO image. The latter does not disappear simply because a dark reference also contains noise.
For a hypothetical night exposure, compare a suspect bright point with another source frame where possible. Keep an actual scene light separate from a sensor-fixed defect. The hot-pixel versus star guide helps frame that first decision.
Do not build a correction library around the assumption that every bright dot should be removed. The reference must describe a camera defect or pattern, not a desired scene feature.
Supply the intended dark-frame directory
In Preferences > Image Processing > Dark-Frame, set the directory containing the reference captures. The documented capture is made with the lens cap on under the same settings, preferably close in time to the affected image.
Use the Dark-Frame panel to select one reference explicitly or enable Auto-Selection. Inspect the reported reference/template information rather than assuming the newest filename will be used.
Automatic matching considers camera identity and exposure metadata. A file's presence in the directory does not establish that it is the reference selected for this image.
Check matching conditions instead of maximizing the number of references
Camera model, ISO, shutter speed and time are relevant to the documented matching process. When equivalent references are available, RawTherapee can average them into a template.
More files do not automatically produce a better correction if they describe incompatible conditions. Organize them so their origin can be identified, and keep unrelated dark-looking photographs outside the calibration set.
The noise-type guide explains why different noise mechanisms need different treatment. Its darktable controls are not RawTherapee dark-frame commands.
Inspect the corrected source before extending the setup
For a proposed acceptance check, compare the known fixed-pattern locations before and after selecting the intended reference. Also inspect smooth shadows and meaningful small lights.
If random noise remains, that alone does not prove dark-frame matching failed. If the target defect remains unchanged, check the selected reference and condition match before adding a stronger unrelated filter.
Keep calibration success separate from artistic cleanup. A local repair to one disputed point may be appropriate, but it is not evidence that a dark-frame correction was measured or that the scene contains no real detail at that position.
Related tasks
- Are the Bright Colored Dots in My Night Photo Stars or Hot Pixels?
- When Should I Use raw denoise Instead of denoise (profiled) on a High-ISO RAW in darktable 5.6?
Sources and scope
- RawPedia: Dark-Frame: Dark frames address fixed-pattern/thermal effects. Automatic selection uses camera, ISO and shutter metadata. Bad-pixel maps require native coordinates and specific naming.
Official RawPedia documentation was opened on October 23, 2026 (Asia/Shanghai). Examples, commands and acceptance procedures are proposed and have not been executed for this guide. Software behavior and paths should be checked against the installed build.
Prepared by Imagild Editorial, the company editorial team responsible for these guides. AI assisted research and drafting. This is documentation-based guidance with proposed examples, not a claim of hands-on testing or an independent product review. Report a correction.
Cover: Pignatta / Pixabay, Pixabay Content License. This is an illustrative source photograph, not an application screenshot or a generated test result.