SeestarFlow

SeestarFlow is a reproducible, non-generative workflow for turning individual ZWO Seestar FITS subframes into auditable deep-sky masters. It preserves the raw files, measures every frame, rejects weak data, stacks with Siril, and records the commands used by both a free processing branch and the production branch.

The project was built around the Seestar S50 Pro, but the ingest and quality stages work with conventional FITS light frames as well.

What problem it solves

Smart-telescope apps make an attractive live stack. Serious reprocessing needs something different: individual FITS frames, immutable originals, measurable quality gates, repeatable integration, and a clear boundary between objective linear processing and subjective finishing.

flowchart LR
    A[Seestar individual FITS] --> B[Immutable ingest + SHA-256]
    B --> C[Frame measurements]
    C --> D{Quality gate}
    D -->|accepted| E[Siril registration + integration]
    D -->|rejected| F[CSV audit trail]
    E --> G[Identical linear master]
    G --> H[Free: GraXpert + Siril]
    G --> I[Production: GraXpert + RC Astro]
    H --> J[TIFF/JPEG + recipe]
    I --> K[Siril + layered Photoshop finish]
    K --> J

Loading

No AI image generation, painted nebulosity, or replacement sky is part of the pipeline. Astronomy-specific machine-learning tools may be used for correction, denoising, or star separation in the optional paid branch; their results must be checked against the source data.

The two processing stacks

Stage	Free stack	Paid stack
Raw archive and QA	SeestarFlow	SeestarFlow
Registration/integration	Siril	Siril
Gradient correction	GraXpert	GraXpert, or GradientXTerminator as an alternative
Color	Siril	Siril plus Photoshop adjustment layers
Optical correction	Siril deconvolution when justified	No deconvolution tool is assumed
Noise reduction	GraXpert	NoiseXTerminator once
Star separation	StarNet/Siril when suitable	StarXTerminator when suitable
Star control	Siril/manual masks	StarShrink only when stars overwhelm the subject
Stretch and finish	Siril	Siril plus a layered 16-bit Photoshop master
Catalog/release	File system	Lightroom, metadata, output pixels, print soft proof

Both branches must start from the same linear master when comparing software. The paid branch only wins when it improves credible detail, stellar profiles, or noise texture without adding artifacts.

Quick start on Windows

Requirements:

• Python 3.11+
• Siril 1.4+
• GraXpert for the free linear-processing branch
• Optional paid production tools: RC Astro, Photoshop, and Lightroom
• PixInsight only for future advanced workflows; it is not required here

git clone https://github.com/Sleepyreaper/SeestarFlow.git
cd SeestarFlow
py -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install -e ".[dev]"
Copy-Item config.example.toml config.toml
python -m seestarflow doctor

Edit config.toml to match the installed executable locations. Then ingest a night of individual FITS files:

python -m seestarflow ingest `
  --source "E:\MyWorks\M31_sub" `
  --target M31 `
  --site backyard `
  --filter broadband `
  --mount eq

python -m seestarflow analyze --session ".\library\M31\<session>"
python -m seestarflow stack --session ".\library\M31\<session>"
python -m seestarflow linear --stack ".\library\M31\<session>\products\<run>\stack_linear.fit"

Optional RC Astro CLI preprocessing with the two licensed tools that support linear FITS (NoiseXTerminator and StarXTerminator):

python -m seestarflow premium `
  --linear ".\library\M31\<session>\products\<run>\linear\gradient_corrected.fits" `
  --kind galaxy `
  --star-separate

Use --dry-run on stack, linear, or premium to inspect generated commands without launching external software.

Use --grax-denoise only for the free branch. The production branch runs GraXpert background extraction without denoising, then uses NoiseXTerminator once. It never stacks two learned denoisers merely because both are installed.

GradientXTerminator and StarShrink are Photoshop plug-ins. They are decision gates in the finishing workflow, not CLI stages. BlurXTerminator is a separate license and is intentionally not assumed by this repository.

Estimate raw FITS storage before a long plan:

python -m seestarflow storage --exposure 10 --hours 8

The default frame size is an early measured S50 Pro telephoto sub. Replace it with Monday's measured size using --frame-bytes.

Hash a final proof, portfolio, print, or archive export into the private release ledger:

python -m seestarflow release --file "D:/Astro/M27-portfolio-v01.jpg" --target M27 --variant portfolio

Repository map

• seestarflow/ — ingest, FITS parsing, quality measurement, orchestration
• scripts/ — portable Siril scripts and Photoshop layered-master builder
• docs/ARCHITECTURE.md — data model and processing boundaries
• docs/OPERATING_PLAYBOOK.md — canonical end-to-end process and tool decisions
• docs/APP_FIELD_CARD.md — literal Seestar app settings and target/filter rules
• docs/INSTALL_WINDOWS.md — complete Windows setup
• docs/CAPTURE.md — how to acquire processable Seestar data
• docs/FREE_VS_PAID.md — fair comparison and tool roles
• docs/BENCHMARKING.md — test methodology and failure criteria
• docs/DATASET_TEST_MATRIX.md — which external datasets are valid proxies, what has been tested, and which missing test is worth acquiring next
• docs/EQ_EXPOSURE_TEST.md — timed EQ/Alt-Az and 10/30/60s experiments, with the capture-compare command for audited equal-integration selections
• docs/PRODUCTION_WORKFLOW.md — the 30-step public workflow and quality gates
• docs/PROCESSING_RECIPES.md — target-class settings for the installed stack
• docs/FIRST_NIGHT_RUNBOOK.md — historical commissioning runbook
• docs/IP_AND_RELEASE.md — provenance, privacy, metadata, and releases
• docs/EQUIPMENT_AND_STORAGE.md — what is required now and before travel
• docs/workflow-sheet.example.json — stage-sheet spec for public process graphics
• tests/ — synthetic FITS and pipeline tests; no astronomy data required

Raw data, integrated masters, previews, and finished images are intentionally excluded from Git.

Important limitations

• Quality thresholds are starting points, not universal truth. Review rejected frames visually before deleting anything.
• The included Siril recipe is a light-frame workflow. Add matched calibration frames when the saved camera data require them.
• Star separation and ML correction are optional tools, not mandatory stages. Dense or undersampled fields can produce false halos or soft residuals.
• Never evaluate processing software using different integration times or independently tuned crops.

License

MIT. Third-party applications and their models retain their own licenses. No third-party datasets or generated image outputs are distributed here.