Seiza 0.12 grew from a plate solver into an imaging engine. The 0.18 series continues that work with saved stacking contexts, more careful calibration, and measurements that help explain how stars vary across a sensor.

As of this post, the core release is 0.18.13. The native Mac and Windows apps have their own version numbers: both are at 0.7.1, bundling the 0.18.8 core.

Pick up where the stack stopped

A saved preview cannot resume an integration by itself. The stacker also needs its registration reference, calibration masters, running statistics, coverage, and rejection state.

Seiza’s LiveStacker can save that working state to a .seiza-stack context and reopen it later. Writes publish by atomic rename after a checksummed payload is complete. The context is processing state; the finished image is still exported separately.

The desktop apps put this into a practical workflow: watch a capture folder, calibrate and integrate arriving FITS or XISF lights, pause, and resume after an app restart. Accepted and rejected counts and an SNR depth chart show how the session is progressing. Display stretching leaves the linear accumulator and exported science data unchanged.

The stacking library documentation covers saved contexts and the APIs for embedding the same engine.

Calibration that explains its decisions

Automatic calibration in the apps can work from raw bias, dark, dark-flat, and flat libraries. Unreadable or mismatched inputs can be set aside with warnings while compatible groups continue. The Windows release documents matching across exposure, temperature, sensor, filter, focal length, and rotator metadata, with reasons for accepted and skipped calibration frames.

That matters when a folder contains several sessions or one bad file. You can inspect what contributed instead of having to guess why a master failed.

Stars as measurements

The apps’ star-analysis action measures the linear source image without a plate solve or catalog download. It reports HFR, FWHM, background, and noise, with a nine-region sensor grid and parallelogram and triangle tilt diagrams.

Version 0.7.1 improves frames whose stars span only a few pixels. If the first detection pass finds too few measured stars, the core retries with more sensitive settings and then native-resolution detection. This can fill regions that previously had too little evidence for a useful diagram.

These are diagnostics of image quality across the field, not an exact measurement of a mechanical adjustment. Compare several similar exposures and account for focus, guiding, seeing, and field curvature before interpreting a pattern as sensor tilt.

On the desktop, in a browser, or in your own code

Seiza for Mac 0.7.1 includes a universal macOS 15+ app, Finder thumbnails, and Quick Look. Seiza for Windows 0.7.1 ships as a signed x64 MSI. Both provide local solving, image browsing, stretching, and calibrated stacking.

For the engine and CLI, use the core downloads, Rust crate, or Python package. To try plate solving in a browser, upload an image at seiza.fyi, powered by seiza-server.

The same engine also powers the stacking and astrometric features in PSF Guard. All the companion repositories are on the Software page.