Flat-Frame Setup Calculator

05How flat frames work — the complete guide

Everything behind the numbers above: what flats do, how to shoot them, and how to read your histogram.

What a flat frame is, and why you need one

A flat frame is a picture of an evenly illuminated blank field — not stars, just uniform light — taken through the exact same optical train as your deep-sky images: same scope, camera, filter, focus and rotation. Because the incoming light is perfectly even, any unevenness in the resulting frame must have been added by your optics and sensor.

That map of unevenness is what calibration software needs. It divides every light frame by a master flat, cancelling the fingerprint of your system: dark corners, dust shadows and gradients. The background goes flat and neutral, so faint nebulosity survives a hard stretch instead of drowning in rings and vignetting.

What flats correct (and what they don’t)

• Vignetting / light falloff — less light reaches the corners, worst in fast systems.
• Dust motes (“donuts”) — out-of-focus rings from dust on filters or windows; smaller spots from dust on the sensor.
• Uneven illumination & internal reflections that create smooth gradients.
• Pixel-to-pixel sensitivity (PRNU) — the fixed pattern of slightly different pixel gains.

The calibration math

Calibration combines your frames like this:

Calibrated = (Light − Dark) ÷ (Flat − FlatDark or Bias) × k

The dark is subtracted (additive signal), the flat is divided in (multiplicative), and k is the flat’s average level so overall brightness is preserved. The key consequence: because the flat divides into every sub, its noise ends up in every frame — which is why a well-exposed, well-stacked flat matters.

What a good flat looks like — the ADU target

Expose into the middle of the sensor’s linear range: roughly one-third to one-half of the histogram. The actual ADU depends on your bit depth:

The formula is ADU = offset + (max − offset) × fraction. A common sweet spot is ~25,000–33,000 ADU on a 16-bit scale. Note that many “16-bit” CMOS cameras actually digitise at 12 or 14 bits and bit-shift up — set the bit-depth field to match whatever scale your capture software measures ADU in.

Why short exposures — but not too short

Short flats (a second or two) mean almost no thermal signal and let you shoot dozens quickly. But there’s a floor:

LED flicker (PWM). Most dimmable LED panels flash on and off hundreds of times a second to control brightness. A very short exposure catches only part of that cycle, producing shifting horizontal banding — worse on rolling-shutter CMOS. Keep PWM-panel flats above about 0.5–1 second (some panels need 2–3 s). The fix for “too bright” is always to dim the panel and lengthen the exposure, never to shorten it. EL and constant-current panels don’t flicker.

Shutter shadow. Cameras with a mechanical shutter cast a moving curtain shadow. At very short speeds that shadow is uneven across the frame, so keep mechanical-shutter flats at or above ~1–2 seconds.

Light sources for flats

A dedicated flat panel is the most repeatable option. Motorised panels double as a dust cover and can be driven directly by N.I.N.A., SGP and Voyager so flats run automatically.

Filter by filter

Shoot — and calibrate — flats separately for every filter. Each filter carries its own dust in its own positions and passes a different amount of light. Narrowband is the big one: an Hα, OIII or SII filter passes only a 3–7 nm slice of the spectrum, so its flats run 5× to 50× longer than luminance — often many seconds even at full brightness. That’s normal. Auto-expose to a fixed ADU target per filter rather than reusing one exposure time.

One-shot colour: measure the brightest channel

On a colour sensor, the red, green and blue photosites see wildly different amounts of light from the same panel — because the panel isn’t spectrally flat and the Bayer filters aren’t equally efficient. Your capture software reports the average of all of them, and that average hides the spread.

We measured a real flat to see how bad it gets. On an ASI294MC Pro shooting a panel through a Takahashi FS-60 at f/5.9, the whole-frame mean was a healthy-looking 43.9% of full scale — but underneath:

Blue sat 2.4× higher than red, and a measurable population of blue pixels had hit the ceiling at 65,534 — while the frame average still looked comfortably mid-range. Those clipped pixels divide badly into every light frame.

So on an OSC camera: split the frame into its Bayer channels and put the brightest one at 40–50%. PixInsight’s SplitCFA, Siril’s channel statistics, or ASTAP will all do it. If your software only gives you a frame average, aim lower than you otherwise would — on the rig above the brightest channel ran about 19% above the frame mean, though that ratio depends entirely on your panel’s colour and your filters, so measure it once for your own setup rather than trusting a rule of thumb.

One consolation: because calibration divides channel by channel, an uneven level across channels is harmless in itself. It only matters because the brightest channel is the one that clips, and clipping is not recoverable.

How many flats, and the combined noise

Stacking N flats reduces their random noise by √N (25 flats → ÷5; 100 → ÷10). Because the master flat divides into every light, that noise doesn’t average away later — so shoot 20–30 per filter for most rigs, and 40–50 for fast optics or narrowband. Beyond ~50 you hit diminishing returns.

Dark flats vs bias frames

Before the flat is divided in, its own pedestal (offset, read noise, and for longer flats a little dark current/amp glow) must be removed with either bias frames (shortest exposure) or dark flats (darks matched to the flat’s exposure, gain, offset and temperature).

• CMOS: prefer dark flats — many CMOS sensors have unstable bias at ultra-short exposures or show amp glow. Stable IMX571-class sensors (ASI2600/QHY268) also work with bias.
• CCD: stable — bias is fine under ~30 s; dark flats for longer flats (Adam Block’s 30-second rule of thumb).
• DSLR: matched dark flats are safest since the sensor is uncooled.

Match everything, and re-shoot when it changes

Keep gain, offset, temperature, binning, filter, focus and rotation identical between flats and lights. Re-shoot flats whenever you rotate the camera, refocus, pick up new dust, or change filters/spacing/reducer. “My flats made it worse” is almost always a train that changed between lights and flats.