AkastroidHow it works
Nine steps between your data and a photograph.You have to do one of them.
This is the whole astrophotography processing workflow, in the order it happens, described honestly - including what each step costs to learn. Akastroid measures your frames and runs the first eight, then hands you the ninth.
Free for sessions up to 25 frames. No account, nothing uploaded.
The two routes
Six programs, or one.
This is not a strawman. It is the chain most people are told to run, and every program in it is genuinely good at its job. The problem is that there are six of them and you have to learn all six before the first one gives you a photograph.
The usual route
Weeks- Capture300 frames, 12 GB, none of which looks like anything
- DeepSkyStacker or Sirilload each set separately, choose a stacking method and a sigma value
- GraXpertbackground extraction, because the sky was brighter on one side
- StarNet or StarXTerminatorseparate the stars so the nebula can be stretched without them
- PixInsightstretch, deconvolve, build masks, recombine - and learn all of it first
- BlurXTerminatora separate purchase, because the built-in deconvolution is not the good one
- Photoshopfinal curves, colour and the crop
- YouTube and the forumsat every step above, for weeks
Somewhere in here, most people stop. Not because the sky was bad - because the eighth tutorial was.
With Akastroid
Three minutes- Drop the folder inlights, and darks/ and flats/ if you shot them
- Press Processevery decision measured from your own frames
- A finished imagethree versions, and a list of what was decided and why
One application, on your own computer, with nothing uploaded. You spend the evening looking at your photograph instead of at a forum thread from 2019.
The long route has the higher ceiling and always will - masks and selective editing are real, and nothing automatic replaces them. That comparison is written out honestly here. This one is about the first image, and about the several thousand pounds of equipment that produces nothing until you get there.
The barrier was never the telescope
A folder of sixty-second exposures already contains the photograph. Every frame in it was collected by an instrument that did its job. What stands between that folder and an image worth printing is not more equipment and not a darker sky - it is a sequence of eight operations, each with parameters, that nobody has explained yet.
That sequence is the actual product of most astrophotography software. The programs are good; what they ask for is a working understanding of signal processing before they will produce anything at all. So the honest version of this page is the workflow itself, step by step, with the part you would otherwise have to learn written next to the part that is now measured.
Step 1
Put the frames in one folder
Lights in a folder, and if you shot them, darks and flats in folders called darks/ and flats/ beside them. Nothing is renamed, converted or exported first.
What this normally costs you: Most tools want each set loaded separately, in a particular order, sometimes in a particular format.
Step 2
Calibrate away what the camera added
Master darks and flats are built and matched to the lights by exposure and dimensions, then subtracted and divided out. Bias is used where it belongs. A missing set is reported once and the session carries on.
What this normally costs you: Knowing which of darks, flats and bias fixes which fault, and that a dark shot at a different exposure or temperature is worse than no dark at all.
Step 3
Throw out the frames that are ruined
Every frame is measured for star size, star shape, star count, drift, background level, gradient strength, noise, signal-to-noise, cloud, satellite and aircraft trails, lost focus and clipping. Each rejection states which of those was wrong.
What this normally costs you: Blinking through two hundred frames by eye, which is slow, inconsistent after the first hour, and blind to the faults that are not visible at screen stretch.
Step 4
Align them on the stars
Frames are registered by matching star patterns, so a field that rotated between exposures still aligns. Sub-pixel accuracy, no reference frame to pick.
What this normally costs you: Choosing a reference frame, setting a star detection threshold, and finding out only after stacking that an alt-az session smeared the corners.
Step 5
Combine them into one image
Integration with the rejection method the session's own numbers ask for - sigma, Winsorized sigma or linear fit - so satellites and cosmic rays are dropped without eating faint signal.
What this normally costs you: Picking a combination method and a sigma threshold, which is a judgement about how many frames you have and how much they vary.
Step 6
Take the sky back out
The light-pollution gradient is modelled from the background and subtracted, and colour is calibrated by matching your field against a catalogue of half a million stars with measured colours.
What this normally costs you: Placing background sample points by hand, and setting white balance by eye - which is where most images acquire the orange or green cast their owner stops being able to see.
Step 7
Stretch it so the faint parts are visible
The black point is set from how much of your frame is actually sky, and the stretch from the histogram in front of it. A wide field of a small nebula and a close crop of a galaxy get different answers.
What this normally costs you: Curves, by hand, repeatedly, with a generic default black point that is roughly three times too high on a wide field and turns the background grey.
Step 8
Sharpen and denoise by measurement
Deconvolution with a point spread function measured from your own stars, and denoising scaled to the measured noise of the stack. Both decline to run when they would do harm.
What this normally costs you: Guessing a radius and a strength, then discovering the ringing around bright stars a week later.
Step 9
Adjust what you want to adjust
Every decision is listed with the number behind it, and every one can be overruled - frames put back in, stretch pulled back, sharpening turned down. Export to JPEG, PNG or 16-bit TIFF, or save the integration unprocessed and finish it somewhere else.
What this normally costs you: This part is the same. It is the only part that was ever supposed to be yours.
Why measuring beats asking
None of the decisions above are matters of taste, and that is the claim the whole approach rests on. Which rejection method suits a stack is a function of how many frames there are and how much they vary, both of which are countable. How hard to deconvolve is a function of measured star width and signal-to-noise. Whether drizzle helps depends on whether the frames were genuinely dithered, which is in the frames. Where the black point belongs depends on what fraction of the image is sky - about 98% on a wide field, where the generic default is roughly three times too high.
Those are measurements somebody is currently making by eye, badly, after months of learning to. Making them properly is not the interesting part of astrophotography. The interesting part is the ninth step, and the night before it.
Nothing is invented
Statistical and learned methods are used to measure your data and choose parameters. No generative model runs, and no detail is painted in. Every feature in the output traces back to a photon your telescope collected, which is the only version of this that is worth publishing under your own name.
Where to go next
- Deep sky - nebulae, galaxies and clusters from FITS or camera RAW.
- Planets - lucky imaging from SER and AVI captures.
- The Moon.
- Image Doctor - the twelve measurements behind step three.
- A first image, start to finish - the same workflow on real data.
Questions about the workflow
Do I need to understand any of this to get an image?
No. That is the point of the page you are reading. Every step above is measured from your frames and carried out without being configured, and the Details tab afterwards tells you what was decided and why. You can learn the workflow from watching it run, which is a better order than learning it before you are allowed a first result.
How long does the whole workflow take?
A hundred-frame deep sky session is two to three minutes on an M-series Mac. A planetary video of eight hundred frames is a few seconds. Changing how the result is finished afterwards is instant, because the stack is not rebuilt.
Is this the same workflow for planets and the Moon?
The shape is the same and two steps differ. Planetary and lunar captures are graded frame by frame for sharpness, aligned in pieces to remove the atmosphere's local warping, corrected for atmospheric dispersion and sharpened with wavelet layers rather than deconvolution. There is no calibration step, because a video capture does not use darks and flats the way a deep sky session does.
What if I already know how to do this?
Then use it as the front half. Akastroid calibrates, grades, registers and integrates in a few minutes, and Save the stacked image, unprocessed writes the integration before any finishing is applied - so you can take it to PixInsight or Photoshop and do the part you actually want to do.
Does any of it happen in the cloud?
None of it. Your frames are read, processed and written on your own machine and nothing about them is ever uploaded. It works with the wifi off.
Try it on your own data
Free for sessions up to 25 frames, which is enough for a real image from a real night. No account, and nothing leaves your computer.
Download Akastroid