From a folder of frames to a finished photograph, with an explanation of what happens at every stage and which of the automatic decisions are worth overriding.

Most astrophotography software assumes you already know the vocabulary. You are asked to pick a rejection algorithm before anyone has explained what rejection is, and to set a midtone transfer function on data you can’t see yet.

Akastroid works the other way round. It measures your data, makes the decisions, and then tells you what it decided and why. This walkthrough follows a real session from folder to finished image and explains what happens underneath at each step. Knowing the decisions is what lets you tell a good result from a lucky one.

Before you start: what to put in the folder

Akastroid reads FITS from any telescope or astro camera, raw files straight off a DSLR or mirrorless body (CR2, CR3, NEF, ARW, DNG and others), and SER or AVI for planetary work. There is no conversion step. Point it at the files your camera wrote.

For a deep-sky session, put your light frames in a folder. If you shot calibration frames, put them in darks, flats and bias subfolders beside the lights, or leave them loose with their IMAGETYP headers intact. Either way they are found automatically. If you have no calibration frames, carry on anyway. A stack of uncalibrated lights with its gradient removed is still a real photograph.

Drop the folder on the window, or use Choose folder. Analysis starts straight away.

Stage one: grading

Before anything is combined, every frame is measured on its own: star count, focus, star roundness, background level and trailing.

What comes back is a table with a score per frame and a reason attached to any rejection, such as star trailing, 2 satellite/aircraft trails or lost focus. Nothing is discarded silently. You can click any row and overrule the decision in either direction.

Two things here are worth reading rather than skipping.

The eccentricity warning tells you whether your stars are oval, and whether that comes from tracking or optics. If it is consistent across the whole session, it is coma, field curvature or camera tilt. No single frame is at fault, so no frames are rejected for it. If it varies from frame to frame, it is tracking, and the bad frames are dropped. That distinction stops you throwing away a whole session over an optical problem you can fix.

The selected count tells you how strict the grading was. Keeping 109 of 134 is normal. If it kept 40 of 134, look at why before going on. The sky may have changed halfway through the session, and you want to know that.

Stage two: two decisions before the long run

The panel before the main button asks at most two things, and both come with their answer already measured.

Drizzle reconstructs at higher resolution, but it only recovers real detail if your frames were dithered, meaning they landed on slightly different sub-pixel positions between exposures. Akastroid measures your frames and tells you, for example, “109 frames dithered across 57.1 px with a sub-pixel spread of 0.91, enough to recover detail below the pixel grid”. Otherwise it disables the option and says “too few frames to drizzle”. This is a property of the data you collected rather than a preference.

Photometric color calibration matches your field against a catalog of 1.48 million stars with measured colors and sets the white balance from what those stars are. It needs the frames to record where the telescope was pointing. A Seestar, a Dwarf or any goto mount writes that. A bare DSLR on a tracker does not, and the option will say so.

If neither applies to your data, both are off and you press the button.

Stage three: what happens when you press it

This is where the quality of the result is decided.

Calibration. Darks, flats and bias are applied where present. Hot and cold pixels are then repaired statistically whether or not you shot darks, because dark subtraction never catches cosmic ray hits and most people have no darks at all. The test is that the optics spread a star across several pixels, while a defect stands alone.

Registration. Stars are matched between frames using triangle patterns, which survive translation, rotation and scale. This matters a great deal on any alt-azimuth mount, including every smart telescope, because the field rotates through the session. Translation-only alignment fails on that data in a way that looks like poor focus rather than an error.

Integration. The rejection method is chosen from two measurements: how many frames you have, and how much your sky level drifted through the night. A deep, steady session gets sigma clipping. A session where transparency changed gets linear-fit clipping, which expects a smooth drift and rejects only what departs from it. A shallow stack gets Winsorized clipping, because estimating a distribution from eight samples and then discarding values throws away real signal.

Gradient removal. A low-order surface is fitted to the sky and subtracted. The overall level is kept rather than driving the background to black.

Color. The background is neutralized with a per-channel shift rather than a scale, because scaling drains the object along with the sky. White balance then comes from star colors, with each star’s local sky measured from an annulus around it rather than from one number for the whole frame. On a field like Orion, where stars sit on bright red nebulosity, that keeps the nebula from being counted as part of the star.

Deconvolution. The point spread function is measured from your own stars, and the strength is set by the signal-to-noise your stack has. It refuses to run in three cases and says which: the image is already sharper than the pixel scale records, it is too noisy to recover more than it amplifies, or the stars are elongated enough that a symmetric model would turn ovals into dumbbells.

Finishing. Stretch, denoise, local contrast and star reduction, with every parameter set from what the image measures.

On a 134-frame DSLR session this takes a few minutes. Planetary video is faster; a 75-frame capture takes about a second of actual stacking.

Stage four: three renders, and which to keep

You get Natural, Enhanced and Dramatic, all from the same data, and you can switch between them instantly. They differ in how hard they stretch, saturation, contrast and star reduction. The measurements behind them are the same.

Enhanced is the default and is right most of the time. Natural is the one to use for anything scientific. Dramatic is for sharing, and it still stays within limits; nothing is invented.

The Stacked / Split / Processed control shows the plain integration beside the finished render. Use it. If the processed version has lost something the stack contained, this is the one place you will see it straight away.

Stage five: Session details, which most people skip

Every decision is written down here: which stacking method and why, how much gradient was removed, whether deconvolution ran, at what strength and against what measured star width, and whether the color calibration succeeded or fell back to a heuristic.

Read it once on your first image. It is the quickest way to learn what happens to your data, and it tells you which of your problems come from processing and which from capture.

What to override, and when

Leave most sliders alone. They are on AUTO, and AUTO shows the value the app measured rather than a generic default. Three are worth touching.

Black point, if the background still looks hazy. The automatic value comes from what fraction of your frame is sky. On a wide field that is 98%, and the target comes down to match.

Star reduction, if you disagree on looks. It is a matter of taste, and the app has no strong opinion on it.

Saturation, for the same reason.

If you find yourself pulling exposure well down on every image, something upstream is wrong. Check the black point first.

Exporting

Maximum quality and Print write uncompressed 16-bit TIFF at native resolution, which is what you want for anything you plan to work on further. A 480×256 planetary crop is around 740 KB and a full-frame DSLR stack is tens of megabytes. Both are lossless; the size difference is pixel count, not quality.

There is also a Save the stacked image, unprocessed button. It writes the integration result before any finishing. That is the plain record of what your frames held, and the right starting point if you would rather finish in PixInsight or Siril.

If the result disappoints

Work through it in this order:

Look at the Stacked view. If the detail is not in the stack, no finishing will recover it, and the problem is capture or registration.
Open Session details and read the log. It says what was decided. A stack that skipped deconvolution because it was too noisy is telling you something about your exposure.
Check the frame table. If a third of your frames were rejected, the session had a problem worth understanding.
Only then reach for a slider.

An experienced imager works through the same order, mostly without thinking about it. Following it is how you avoid adjusting sliders at random.

Your First Image in Akastroid: A Complete Walkthrough