A good flat corrects repeatable variations in how light reaches different parts of the sensor.
Typical examples include vignetting, dust shadows, and pixel-response differences.
That means a flat is not a generic “make the image better” frame.
It is a measurement of a particular optical configuration.
When applying flats makes the data worse, describe the new artifact before changing the whole preprocessing workflow.
Start with the pattern the flat created
Different outputs point toward different failure families.
Corners become too bright: the vignetting correction is being over-applied.
Dark dust donuts remain: the flat does not reproduce the same dust pattern well enough.
Dust donuts become bright rings: the correction is strongly mismatched or over-applied.
A new gradient appears: investigate the flat illumination or another light-source problem.
One filter calibrates well and another does not: investigate filter-specific flats, exposure, and optical geometry.
The flat looks plausible but calibrated lights are worse: inspect how the flat itself was calibrated before assuming the optical pattern is wrong.
The output is evidence.
Confirm that the flat matches the light-frame geometry
The flat needs to represent the same relevant optical path.
Check whether anything changed between lights and flats:
- camera rotation;
- filter;
- reducer, flattener, or other focal-length-changing optic;
- major focus position;
- filter wheel or drawer position;
- imaging-train assembly;
- dust position.
If a dust shadow moves on the sensor, the old flat cannot remove it from the new position.
It may leave the original dark mark and add a false bright correction elsewhere.
Do not dismantle and rebuild the imaging train before flats if you expect the dust and illumination pattern to remain matched.
Do not assume one master flat fits every filter
Filters can differ in transmission, illumination behavior, dust position, and mechanical seating.
Test whether one master is genuinely valid across filters rather than assuming it.
A repeatable failure on only one filter is valuable evidence because much of the rest of the imaging system is shared.
Check the flat illumination source
The source does not need to be perfect in an abstract sense, but it must not add a strong gradient unrelated to the optical system.
If calibrated data develops a new gradient, compare test flats while changing the illumination source position without changing the camera.
If the brightness pattern follows the source, the source is suspect.
A flat should measure the telescope/camera illumination pattern, not the unevenness of the panel setup.
Check signal without clipping
A useful flat needs enough signal to measure response accurately while staying below saturation.
A very dark flat has poor signal relative to noise.
A clipped flat no longer preserves proportional response in the saturated pixels.
Inspect the histogram or pixel statistics rather than relying on how bright the flat looks on screen.
Some camera and illumination combinations also behave poorly at extremely short exposures. If you see banding or inconsistent brightness, test a longer exposure with a dimmer source and compare repeatability.
Keep light darks and flat darks conceptually separate
A light dark is intended to match the exposure characteristics of the light frame.
A flat dark, when used by the camera-specific workflow, matches the flat exposure.
They solve different calibration tasks.
Do not mix calibration masters simply because they came from the same camera.
Match exposure, temperature where relevant, gain, offset, and readout mode to the workflow being used.
Audit the zero-level treatment
Raw frames contain an electronic baseline or offset.
If a master flat is normalized without accounting for that baseline correctly, the multiplicative correction can be wrong even when the optical pattern looks right.
This is one reason apparently sensible flats can create bright corners or inverse dust patterns.
Depending on the camera and processing workflow, flat calibration may use a master bias, a synthetic offset approach, or matching flat darks.
There is no single universal frame recipe for every sensor.
Use a method appropriate to the camera and software, then keep it internally consistent.
Find the first calibration stage that creates the defect
Do not recalibrate an entire night after every guess.
Use one representative light and compare stages.
For example:
- inspect the raw light;
- inspect it after the relevant dark correction;
- apply the current master flat;
- rebuild the flat correctly if the evidence points there;
- compare again.
The useful question is:
At which stage does the bad pattern first appear?
That is the stage to investigate.
Do not hide a broken calibration chain in post-processing
Background extraction can hide broad residual gradients.
Cosmetic correction can hide hot pixels.
Retouching can hide dust.
None of those proves the underlying calibration is correct.
Where practical, make the calibrated sub technically sensible before aggressive post-processing begins.
Once one representative light calibrates cleanly, verify the correction on a small set before processing the entire session.
For star-shape failures that remain after calibration, return to the trailing-star and optical-geometry checks rather than treating every image defect as a flat problem.