How Many Frames Do You Need? A Straight Answer

“Shoot as many as you can” is the standard advice, and it is not so much wrong as useless. It gives you no way to plan a session, no way to know when to stop, and no way to understand why last week’s two hours produced something noticeably better than this week’s three.

There is a straight answer, and it follows from one relationship.

The arithmetic

Signal accumulates linearly with the number of frames. Noise, being random, partly cancels and grows only with the square root. So the ratio improves by √N.

That single fact has consequences most people find surprising:

Frames	Improvement over one frame
4	2×
16	4×
64	8×
256	16×
1024	32×

Every doubling of quality costs four times as many frames.

This is why 10 frames to 40 is transformative and 200 to 260 is invisible. You added 30% more data for a 14% improvement, and 14% sits below what the eye reliably notices in a photograph. Nothing went wrong. You bought a quantity of improvement that does not register.

The practical rule: to see a real step up, plan to quadruple, not to add a bit. If 50 frames gave you something decent, 200 is the next visible level. 70 is not.

Read noise decides your sub length

There is a second noise source that does not average away the same way. Every sensor readout adds a fixed penalty. It does not depend on exposure length — a 5-second frame and a 500-second frame each pay it exactly once.

So a given total integration carries read noise in proportion to how many pieces you cut it into:

• 90 minutes as 540 × 10-second subs → 540 helpings of read noise
• 90 minutes as 18 × 5-minute subs → 18 helpings

Same total light. Roughly thirty times less readout penalty.

This is the entire reason a guided setup pulls ahead of a smart telescope on faint targets at identical total time. It is not collecting more photons. It reads the sensor far less often.

The rule: use the longest sub your mount reliably tracks. That is a mechanical question, not a processing one, and it dominates everything else on faint targets.

The counter-pressure is obvious — a trailed 300-second frame is worth nothing, and a lost frame costs more than a slightly noisier one. Find the length where your keep rate stays high, then stay there.

What “enough” looks like

Rough working numbers, assuming reasonable skies:

Bright targets — Orion’s core, the Pleiades, Andromeda’s centre, the Moon. Twenty to fifty frames gets a genuinely good result. These are bright enough that you are not fighting the noise floor, and more time yields little.

Medium targets — most Messier galaxies and nebulae. One hundred to three hundred. This is where the √N curve does real work and each additional hour still shows.

Faint targets — outer nebulosity, integrated flux nebula, faint tidal tails. Five hundred and up, though honestly this is where sub length and sky quality matter more than count. If you are read-noise limited or sky limited, frames alone will not rescue it.

Planetary — a completely different regime. You want thousands, because you are throwing most away. Lucky imaging keeps the sharpest fraction, so 5,000 frames at 10% leaves 500 good ones, and it is the selection doing the work rather than the integration.

Three limits, and how to tell them apart

Adding frames only helps if frames are what limits you. Three things commonly are not:

You are read-noise limited. With very short subs the noise floor is set by readout, not by sky photons. More frames still help, but slowly, and you pay a fixed cost repeatedly. Symptom: long integration, stubbornly noisy background, short subs. Fix: longer subs, not more of them.

You are sky limited. Under bright skies the background dominates and faint signal may never rise above it. Signal-to-noise still improves as √N — but √N of “undetectable” is still undetectable. Symptom: bright background, faint regions that stay absent no matter the integration. Fix: darker skies, or narrowband filters that reject most light pollution.

Your alignment is limiting. If registration leaves half-pixel errors, more frames add more slightly-misregistered data. The stack gets smoother and no sharper. Symptom: noise falls with more frames but detail never improves, and stars look soft in a way that does not match your focus. Fix: check that registration handles rotation — which matters on every alt-azimuth mount, including all smart telescopes.

That last one deserves checking before you blame the sky, because it looks exactly like poor seeing and people lose seasons to it.

The diagnostic that settles it in one run

Stack your session. Then stack half of it — every other frame — and compare the two at the same stretch.

• Clearly better with all of them? Frames are helping. Shoot more.
• Barely distinguishable? You have hit one of the three limits above. More frames will not fix it; change something else.

This costs one extra processing run and tells you more than any rule of thumb, because it is measured on your data, your sky and your equipment rather than on someone else’s.

Rejection needs a minimum

One thing that genuinely does depend on raw count: outlier rejection.

With eight frames you cannot estimate a distribution. Any clipping method is working from too few samples, and aggressive rejection discards real signal while claiming to remove outliers. The failure mode is subtle — the image looks clean and is quietly missing faint detail. It does not look broken. It looks slightly empty, and nothing on screen explains why.

Practical thresholds:

• Under 8 frames: plain averaging or a median. Accept the satellite trails; you cannot safely reject them.
• 8 to 15: Winsorized clipping, which pulls outliers toward the bulk rather than discarding them.
• Above 15: ordinary sigma clipping becomes safe.
• Above 15 with drifting transparency: linear-fit clipping, which expects a smooth drift across the stack and rejects only what departs from it.

That last case is worth knowing. When high cloud rolls in or your target sinks toward the light dome, the frames genuinely differ. Sigma clipping treats frames at the ends of the range as outliers — and the darkest ones are often the clearest frames of the night. You lose your best data to a method that assumed it was wrong.

So: how many?

• Sub length first. The longest your mount tracks reliably. This dominates read noise, and read noise dominates faint targets.
• Then count, in factors of four. 50 → 200 → 800. Adding 20% is not a plan.
• At least 15 frames if you want proper outlier rejection.
• Verify with the half-stack test rather than trusting any of the above, this article included.

And when the half-stack looks the same as the full one, your next clear night is better spent on a different target than on more of the same one.