How Many Frames Should a Planetary Stack Keep?

Every planetary tutorial gives a number. Keep the best 25%. Keep the best 10%. Keep 500 frames. The numbers disagree because each was right for the capture its author had in front of them, and nobody says which kind of capture that was.

The right fraction is a property of your capture, specifically of how quickly frame quality falls off from the best frame to the median one, and you can measure it rather than guess. Below is that measurement on two real captures. They came out differently, which is why both are here.

What you are trading

Lucky imaging works because the atmosphere is not equally bad at every instant. Keeping only the sharpest frames buys resolution. Averaging more frames buys signal-to-noise. The two pull in opposite directions. Every frame you add past the sharpest few is a little softer than the ones before it, and averaging a soft frame into a sharp stack does not average away. It low-passes the result, the same as a blur would.

So there is a knee. Above it, adding frames costs more sharpness than it returns in noise. Below it, the stack is grainy and wavelet sharpening amplifies the grain. Where the knee sits depends on the shape of the quality curve, and the seeing sets that shape.

Two captures, graded

Akastroid grades every frame on fine-scale structure relative to the frame’s own noise, so a noisy frame does not score as sharp, and normalises the scores so the best frame is 100. This is where the top of each capture’s ranking sat:

Fraction kept	Jupiter, steady air (11,473 frames)	Saturn, rough air (17,934 frames)
top 1%	91	72
top 5%	85	62
top 10%	82	56
top 25%	75	48
median frame	65	40

On the Jupiter night, the 25th-percentile frame scored 75, three quarters as sharp as the best moment of the evening. On the Saturn night it scored 48, and the median frame 40. A “best 25%” stack of that Saturn averages in frames less than half as sharp as the best ones.

The stacks

Each capture was stacked at 25%, 12% and 6% with everything else identical. The results were measured for edge energy (how much fine detail survives at the belt and ring boundaries) and for residual noise on the disc.

Jupiter: no difference. Edge energy and noise were the same within measurement error at all three fractions. When the top quarter of the capture is nearly as good as the top frame, it hardly matters where you cut.

Saturn:

kept	frames	edge detail	noise
25%	4,484	0.38	1.50
12%	2,152	0.41 (+8%)	1.56 (+4%)
6%	1,076	0.45 (+18%)	1.73 (+15%)

Going from a quarter to an eighth bought 8% more edge detail for 4% more noise. That trade is worth taking, and the noise increase is smaller than what sharpening adds anyway. Going on to a sixteenth bought more detail again, but the noise cost starts to show on the globe. The knee for that capture is around 12%, nowhere near 25%.

What Akastroid does with this: it stopped cutting

This guide used to end with a rule for picking the fraction. The rule was measured, it was wrong, and it has been removed.

The question started with a Jupiter that looked blurry. It had been stacked from 103 of 1,200 frames, 8.6%, which is what the conventional advice asks for. So each capture was stacked at its own recommended fraction and again at 100%, everything else identical, and judged on the finished master’s limb width (which noise cannot flatter) and on sky noise:

capture	recommended	limb, cut → all	noise, cut → all
Jupiter SER	8.6%	35.5 → 34.0 px	0.00011 → 0.00006
Saturn AVI	16.4%	41.5 → 42.5 px	0.00005 → 0.00004
Jupiter AVI	19.6%	45.3 → 45.3 px	0.00015 → 0.00010

Stacking everything was sharper on one capture, identical on one, and half a camera pixel softer on one. It was cleaner on all three, by up to 45%.

That runs against what the hobby teaches, but the reason is simple. The frames you throw away are slightly softer, not worthless. Averaging them in costs a little sharpness and buys a lot of noise reduction, and once multi-point alignment is doing its job, that trade favours keeping them. A cut in front of the aligner throws away signal to guard against something the aligner already handles.

So Akastroid no longer cuts to the conventional few percent. It keeps every frame that scores close to the best, and never fewer than half the capture; on most captures that is exactly half. The word “recommended” was taken out of the interface. The quality curve and the drag handle are still there: drag to 100% and the button reads STACK ALL FRAMES, or cut harder when a capture has real cloud or a bumped tripod in it, which you can see on the graph.

How to read your own capture

The frame browser shows the graded frames in order. Scrub from the best frame toward the median and watch the belts. On a steady night the first few hundred look alike and the softening is gradual. On a rough night the drop is visible within the first hundred, and by the median frame the planet is a smear.

That is the quality curve, seen rather than plotted. It shows how deep into the capture the luck goes.

The short version

• The conventional 5-25% rule was measured against stacking everything on three real captures. Stacking everything was sharper on one, level on one and marginally softer on one, and cleaner on all three.
• The frames you discard are slightly soft, not worthless. Averaging them in trades a little sharpness for a lot less noise.
• Akastroid keeps the frames that score close to the best and never fewer than half. The handle goes up to every frame, or lower for a capture with real cloud in it.
• If your planet looks blurry, first check how few frames went into it.