A folder of subs in.A finished image out.

The problem with a folder of subs

Deep sky imaging produces an amount of data that is out of all proportion to the picture at the end of it. Three hundred sixty-second exposures is five hours of sky and about twelve gigabytes of files, none of which looks like anything on its own. The image is in there, spread thinly across every frame, and getting it out means calibrating each one, throwing away the ruined ones, aligning the rest to sub-pixel accuracy, combining them in a way that rejects satellites without eating faint signal, removing whatever gradient the sky put across the frame, fixing the colour, and stretching the result without turning the background grey or blowing out the core.

Every one of those steps has a parameter. Most programs ask you for all of them before they will do anything, which is reasonable if you already know the answers and an impasse if you do not. That is the actual barrier to a first good image - not the telescope, not the sky, but a sequence of decisions nobody has explained yet.

Akastroid measures the session and answers them from your data. It is not a preset. A wide field of a small nebula and a close crop of a galaxy get different answers because they measure differently.

Calibration, found rather than configured

Put your darks in a folder called darks/ and your flats in flats/ beside the lights and that is the whole setup. Akastroid builds the masters, matches them to the lights by exposure and dimensions, and applies them. Bias frames are used where they belong. If a set is missing, the session is stacked without it and the log says which one and what it would have improved - a first session with no calibration frames at all still produces an image, which matters more than being correct about best practice at somebody who has just come in from the garden.

Camera RAW is read directly, with the black level removed and nothing else done to it - no demosaic guesswork from a third-party converter, no gamma quietly applied. FITS keeps its header: Bayer pattern, exposure, gain, filter and pointing coordinates are all read and used. A Seestar folder works with no configuration because everything needed is already in the files it writes.

Grading: which frames are worth keeping

Every frame is measured before anything is stacked. Star count, sharpness, roundness and how far each frame's stars have drifted from the session's own norm; background level, gradient strength, noise and signal-to-noise; and named artefacts - cloud, satellite and aircraft trails, lost focus, overexposure.

Two things about that list matter more than the list itself. The first is that every rejection states its reason, so the frame table is a description of your night rather than a verdict - you can see the hour when the transparency dropped. The second is that the norm is the session's own. Oval stars in every frame are a property of the optics or the mount, not of any individual frame, and Akastroid reports that as a session-level observation instead of rejecting the entire night for being consistently imperfect.

Any grade can be overruled with a click. A frame you know is fine goes back in.

Registration that survives field rotation

Frames are aligned by matching star patterns - asterisms - rather than by assuming one frame is a shifted copy of another. That distinction is what decides whether an alt-az session stacks at all. A Seestar, a Dobsonian or any undriven mount rotates the field slightly between exposures, and a stacker that only searches for translation produces progressively worse alignment towards the frame edges, which shows up as stars that streak into arcs away from the centre.

Where your frames were dithered between exposures, drizzle reconstructs at higher resolution than the sensor sampled. Akastroid measures whether the dither actually happened and says so, rather than offering a control that quietly does nothing on undithered data. Multi-panel sessions are matched into an overlap graph, level-matched at the seams and feathered together, so a mosaic is a session rather than a project.

Integration chosen from the stack you actually have

Averaging, median, sigma clipping and winsorized clipping are all available, and the choice is made from depth and behaviour rather than from a preference. A shallow stack cannot support aggressive rejection - throw out the outliers from eight frames and there is not enough left to average. A night whose transparency drifted needs a method that expects the drift. Both are read from your frames.

The decision, and the measurement behind it, are written into the log with everything else. If you disagree, change it: the stack is rebuilt, and the finishing that follows is unaffected.

Gradients, colour and the stretch

Light pollution, moon glow and vignetting are modelled and removed while the sky keeps its level. That last part is the difference between a photograph and a poster: a background subtracted all the way to black reads as a cutout, because a real sky is not black and the eye knows it.

Colour is measured, not eyeballed. The field is matched against a catalogue of stars whose colours were measured by Tycho-2, and the channel weights are fitted so that those stars come out the colour they actually are. This is photometric colour calibration, the step most guides recommend and most people skip because setting it up is fiddly, and here it happens by default. A nebula that came out magenta because the white balance was guessed comes out the colour it is.

The stretch curve is solved from your own histogram, with bright cores protected so the heart of M42 stays a structure rather than becoming a white blob. Background level is set from what fraction of the frame is sky - often 98% on a wide field, where a target chosen for a half-full frame would lift the whole background to grey. Deconvolution measures the blur from the stars in your own image, and declines outright when the signal-to-noise cannot support it, because sharpening noise faster than detail is how a stack gets worse.

Three finished versions, then the file you need

Every session comes back as Natural, Enhanced and Dramatic, rendered from the same integration. Switching between them is instant. From there the colour tools are conventional and live: histogram, curves, hue-selective saturation, green-cast removal, temperature and tint.

Export is JPEG and PNG for sharing, 16-bit TIFF for printing, and a FITS master that keeps the float data along with a record of what was done to it. There is also Save the stacked image, unprocessed, which writes the integration before any finishing - so Akastroid can be the front half of a workflow that ends in PixInsight or Photoshop. A hundred-frame session takes two to three minutes, which makes it affordable to find out whether last night is worth an evening of manual work.

Where it fits, honestly

Siril is free and excellent, and in expert hands it will go further than this. PixInsight goes further still and has a learning curve to match. Neither is trying to answer the question Akastroid answers, which is what to do when you have the data and not the vocabulary - or the evening.

If you already know which rejection algorithm your night needs and enjoy choosing it, you do not need this. If you would rather look at the picture, or you have four targets to get through before work, this is the case it was built for. The comparison on the pricing page is laid out in full, including the parts where the free tools win.

Deep sky guides

The processing itself is explained in the guides, independently of the software: how stacking works, what darks, flats and bias actually do, how many frames you need, removing light pollution gradients, why your nebula came out red and why your stars are trailing. If you shoot with a smart telescope, the Seestar FITS guide covers what that device writes and what to do with it.