A week ago I wrote about my Seestar pipeline going live: my Windows PC stacks the telescope’s frames in Siril, and the Raspberry Pi uploads the result to Flickr and drafts a post on my astronomy site, vividgrey.co.uk. At the time it had produced two images, the Whirlpool Galaxy and the Pacman Nebula.
It’s now been running for two weeks, so this is a short follow-up on what it has actually done, rather than what it was designed to do. The short version: five finished images, four of them published, and very little effort on my part.
The first fortnight in numbers
Each run on the PC processes one target, picking the deep-sky object with the most unprocessed data. Here is what came out:
- M 51, the Whirlpool Galaxy (8 November): 4,628 frames, 12 hours 52 minutes, published 9 November.
- NGC 281, the Pacman Nebula (10 November): 2,002 frames, 5 hours 34 minutes, published 12 November.
- IC 434, the Horsehead Nebula region (14 November): 1,617 frames, 4 hours 29 minutes, published 15 November.
- M 81, Bode’s Galaxy (17 November): 1,149 frames, 3 hours 11 minutes, published 19 November.
- M 13, the Hercules Cluster (20 November): 825 frames, 2 hours 17 minutes, waiting for me to review.
That’s 10,221 ten-second frames, a little over 30 hours of light, all taken with the Seestar S50 between April 2025 and April 2026 and sitting on the NAS until now. Every image was colour-calibrated with Siril’s SPCC against the Gaia star catalogue, using anything from 115 stars (M 51) to 1,239 (NGC 281).
The new arrivals
The first two images were in the last post, so here are the three that have arrived since.

IC 434 is the glowing cloud in Orion that the dark Horsehead Nebula stands out against. The Seestar had captured 2,216 frames over nine nights between December 2025 and March 2026. 1,806 of them aligned and 1,617 made it through the filters into the final stack, calibrated against 232 stars. The horse’s head is small in the Seestar’s field of view, but it’s unmistakable.

M 81, a spiral galaxy in Ursa Major, shows one of the pipeline’s limits. My notes list ten nights on it, but on eight of those the Seestar only kept its own stacks, not the individual frames, so Siril could only use two nights: 1,149 of 1,325 frames, calibrated against 184 stars. The background also isn’t perfectly even, which is something to look at in the processing later.

M 13 is a globular cluster, a dense ball of stars rather than a galaxy or a nebula, and the first of its kind through the pipeline. 825 of 1,074 frames from three nights in 2025 were stacked, calibrated against 339 stars. Its vividgrey post is still a draft while I look at it.
Twice a week, and a missed Friday
The PC job started as Friday only. On 9 November, the day after the first run, I added a Tuesday run as well, so images arrive twice a week. The Pi didn’t need a new schedule, because it already checks for new images every morning at 07:30. Its “no new image” check, which used to run only on Saturdays, now also runs on Wednesdays, and the message names the run it was expecting.
That check has fired once so far. On Saturday 14 November my Daily note said there was no new image from the PC’s Friday run and asked whether the PC had been switched on. IC 434 was then processed later that Saturday, and the Pi picked it up on Sunday morning. Since then the Tuesday and Friday runs have both delivered on time.
The PC side runs only while I’m logged on, because it needs the vault and the NAS, and the scheduled task is set to run as soon as possible after a missed start. That’s the price of doing the heavy work on a desktop rather than a server, and so far it’s a price I’m happy to pay.
The review reminder
Nothing goes public on vividgrey until I press Publish, so the pipeline has to make sure I actually do. When the Pi creates a draft, it adds a checkbox to that day’s Daily note telling me to review and publish it, with edit and preview links. If the post is still a draft a week later, the reminder comes back, and it keeps coming back every seven days until I publish. Once I have, the next run marks the target’s note as published, with the date and the URL.
So far the repeat hasn’t been needed: each of the first four drafts was published within a day of appearing. Having the to-do sitting in the note I open every morning turns out to be most of the battle.
Faster than expected
In the last post I said that copying the frames from the NAS was the real bottleneck: for M 51 it took 84 minutes, against 207 minutes in Siril. M 51 turned out to be the extreme case, because it has by far the most data. The later targets were much quicker:
- NGC 281: 16 minutes copying, 122 minutes stacking.
- IC 434: 14 minutes copying, 99 minutes stacking.
- M 81: about a minute copying, 39 minutes stacking.
- M 13: about 2 minutes copying, 14 minutes stacking.
Because the job always takes the target with the most data first, each run should be a bit quicker than the one before, at least until I capture something new.
Next: the Moon, the Sun and Jupiter
Until now the pipeline only handled deep-sky objects. The Seestar records the Moon, the Sun and the planets as video rather than as individual frames, and Siril’s alignment only works with stars, so it can’t stack them from a script.
On 21 November Claude Code added a separate “lucky imaging” step on the PC, written in Python with OpenCV. It scores every frame of a video for sharpness, keeps the best 10%, aligns them, averages them and sharpens the result. In a test on one of my Moon videos it kept 410 of 4,100 frames, and the stack was clearly sharper than a typical single frame. The Pi side was updated the same evening to build the right Flickr description and post for a video (capture date, video length, frames kept) instead of the deep-sky details.
- One video per run. Each Tuesday and Friday run now does its deep-sky target first and then one video, so the deep-sky queue isn’t held up.
- Ten videos waiting. Four of the Moon, two of the Sun and four of Jupiter. The largest goes first, which should be a Moon video from August 2025.
- Turned on carefully. The video step was switched off until the Pi could handle its output, because otherwise the publisher would have failed on it. It was only switched on once that change was in.
The first solar-system image should arrive after Tuesday’s run. I’ll find out on Wednesday morning whether it worked.
What I’ve learned
- Running it beats designing it. Two weeks of real runs told me more about copy times, missing frames and missed schedules than any amount of planning would have.
- The checks earn their keep. The “no new image” message caught the missed Friday the first time it could have.
- A reminder in the right place is enough. I didn’t need anything clever to get drafts published, just a checkbox in the note I already read every day.
- Old data is worth revisiting. Every one of these images came from frames I’d captured months ago and never done anything with.
The pile isn’t gone yet, but it’s shrinking twice a week.
A note on how this post was written: it was drafted by Claude Code from the AstroProcessor and AstroPublisher repos, their code history and my vault notes on the pipeline and each target, and written into WordPress as a draft through the API. Claude chose the photos from my Flickr account. I reviewed and edited it before publishing.
Leave a Reply