Measured beside a desktop star-trail app: speed, memory and line removal

Startrails Stacker 2.1 is a separate, paid desktop application for Windows and macOS, with a name close to this site's. On 23 September 2026 its trial build and this tool were run side by side on one Windows laptop, each through its own controls, on two public star sequences and on the site owner's own camera-RAW night. On the two public sequences with no line removal this tool finished first and used about a quarter of the memory. On 30 camera-RAW frames the desktop app was seven times faster. With line removal on this tool was slower, but it removed the labelled aircraft, and the desktop app, at its default strength and at its maximum, left them in. Measured again two days later, after seven changes that each left its results byte for byte the same, this tool took 39.2 s on those RAW frames, against the desktop app's 16.4 s.

How it was measured

Both tools ran on the same laptop: an AMD Ryzen 9 6900HS with 16 logical processors and 47 GB of memory, on Windows 11. The desktop app was operated through its own window, setting and then reading back every control before each run. This tool was operated through its own page in Chromium, on this site's build of that day. Time runs from the click on the start button to the finished result; memory is the peak of each tool's whole process tree, sampled every 100 ms. Each case ran once to warm up and then three more times, alternating which tool went first, and every figure below is the median of those three. Other virtual machines were running on the same laptop, so every time is a little slower than an idle machine would give.

The inputs were two public sequences and one private night: 16 all-sky frames from Zenodo record 3736793 (CC BY 4.0), prepared as 16-bit PNG; ten frames of the Yebes Observatory planes-and-meteors night (image property of Yebes Observatory), with five aircraft positions labelled in advance; and the first 30 frames of the site owner's own Sony A7C night, 24-megapixel ARW files, run through the desktop app with the owner's permission.

What was measured

On the 16 all-sky frames with no line removal, this tool took 6.6 s and 0.48 GB, and the desktop app took 9.1 s and 2.07 GB. On the ten Yebes frames with no line removal, this tool took 6.8 s and the desktop app 12.1 s. The desktop app's memory was not steady: the same case peaked anywhere from 0.34 GB to 2.19 GB from one run to the next. Its time hardly depended on the number of frames: three black frames took it 11.3 s, and it ran as 13 processes.

On the 30 Sony A7C RAW frames with no line removal, the desktop app took 16.4 s and this tool 122 s; both peaked near 2.4 GB. Almost all of this tool's time went to building the result frame by frame, about 3.3 s per frame. With its default line removal on, the same 30 RAW frames took this tool 313 s and 2.32 GB; the desktop app's default has removal off. The desktop app wrote a full-size 16-bit TIFF of the night and this tool a full-size 16-bit PNG.

Measured again on 24 and 25 September

The next day this tool had three changes that made it faster, and each one left its results byte for byte the same. The same driver ran it again on the same laptop, once to warm up and then three more times per case, and every figure below is again the median of those three. On 25 September, after four more such changes, the same driver ran it once more in the same way. The desktop app was not run again, so its figures are still the ones from 23 September. The laptop was about as busy as the day before: 20% to 88% of its processor was in use before a run, against 20% to 90% then. On 25 September it was 31% to 56%.

On the 30 RAW frames with no line removal this tool took 75.2 s against the desktop app's 16.4 s, so the desktop app was about four and a half times faster instead of seven. Both RAW results were the same files as the day before. With line removal on the ten Yebes frames this tool took 31.2 s against the desktop app's 12.7 s at its default strength, about two and a half times as long, and the four labelled aircraft still kept a median of 8% of their contrast. On 25 September the RAW frames with no line removal took 39.2 s against the desktop app's 16.4 s, and the ten Yebes frames with line removal on took 24.2 s against its 12.7 s. Every result on 25 September was the same file as on 24 September.

Line removal

With line removal on, this tool took 39.9 s on the ten Yebes frames and the desktop app 12.7 s at its default strength and 13.5 s at its maximum. On the four labelled aircraft trails that could be scored, this tool left a median of 8% of each trail's contrast against the sky beside it; the desktop app left all of it, at its default strength and at its maximum. Scoring compares each tool's result with removal on against the same tool's result with removal off, so the two tools' different picture tones cannot count as a difference in removal. The desktop app's slider runs from 1 to 100 and starts at 1; at 100 the pixels it changed were at the horizon and the edges of the frame, in ground lights, clouds and the printed caption, not on the aircraft.

Without an exposure time or an image scale this tool will not search for lines, because a star trail and an aircraft cannot then be told apart. Video stills like the Yebes frames carry none, so its default run left the aircraft in and said why, until the exposure and interval were typed into the page. The desktop app has removal off by default.

Where this tool was weaker

• On camera RAW it was seven times slower than the desktop app. Since this comparison, it takes 39.2 s where the desktop app takes 16.4 s.
• With line removal on it was about three times slower than the desktop app on the same ten frames. Since this comparison, it takes 24.2 s where the desktop app takes 12.7 s.
• It needs the exposure time of frames that do not record it before it will remove anything.
• Five of the 13 lines it removed on the Yebes night were edges of bright clouds moving along the horizon, not aircraft; it kept the ground, but it counted clouds as sky. Since this comparison, a broad line running along the ground is offered for review instead of removed, which keeps those five and still removes the four aircraft.
• On the compressed Yebes video stills its per-frame sky levelling left faint coloured blocks in the dark sky that a strong stretch shows. Since this comparison, it finds a video compressor's blocks in the result and evens out the colour of the dark sky around them, keeping every pixel's brightness; the brightness blocks remain, and so does some colour beside the brightest stars.

What this does not show

• The desktop app's features page says 405 full-resolution RAW images of 6336 × 9504 pixels are stacked in just over a minute; no files of that size were run here, only the 24-megapixel A7C night.
• Only a trial build was run. It writes a large diagonal “Trial Version” mark across every result; those pixels were left out of every image measurement for both tools, and a licensed build was not tested.
• One laptop and one day; not macOS, and not the desktop app's comet, timelapse or chunk features.