Free and open source · Windows, Linux, macOS · alpha

Astraios — Free GPU-Accelerated Astrophotography Processing

Astraios is a free, open-source desktop application for calibrating, registering, stacking and processing deep-sky, planetary and solar astrophotography on Windows, Linux and macOS. Supported operations use NVIDIA CUDA or Apple Metal acceleration with automatic CPU fallback. Image processing and AI inference run locally; explicitly selected online integrations such as astrometry.net transmit only the data that service needs. Astraios is currently alpha software.

Download v0.1.25-alpha View on GitHub Install guide

GPU where it mattersCUDA and Apple Metal for registration, stacking, deconvolution, denoising, stretching and AI; automatic CPU fallback on every machine.
Local-firstNo account, no telemetry, no licence key. The only image upload is the optional astrometry.net solver.
GuidedA step-by-step wizard that suggests settings from your own image and explains every control.
Honest about maturityAlpha. Complete workflows, rough edges, one maintainer. Bug reports are welcome.

Astraios 0.1.25 main window: project panel on the left, a synthetic star field with a ring nebula on the canvas, the Stretch tools with Auto-Stretch open on the right, and the processing log reporting CUDA on an RTX 5060. — The interface with the Stretch tools open. The image on the canvas is a synthetic test field generated by the project's screenshot script, not a real astrophotograph.

How a session goes

Load. Lights, darks, flats and bias frames as FITS, XISF, TIFF, PNG or JPEG; SER video for planetary work.
Calibrate and cull. Master frames, cosmetic correction, subframe scoring by FWHM, eccentricity, SNR and star count.
Register and stack. Star or triangle matching on the GPU, winsorized or sigma rejection, optional drizzle, multi-session and live stacking.
Guided Processing. Crop, gradient removal, colour calibration (PCC, SPCC, SFCC), deconvolution, denoise, stretch and saturation, each previewed and explained.
Finish and export. Narrowband palettes, star removal and recombination, layers, annotation; export to FITS, XISF, TIFF, PNG or JPEG.

What sets it apart

Built in, not bolted on

Spatially varying and multi-frame deconvolution, AI denoise with a bundled model, star removal and photometric colour calibration are part of the program, not paid plugins.

Verified registration

Every frame's transform is checked before it is accepted; frames that cannot be registered are left out rather than stacked unaligned, and registration padding never darkens the edges.

Science tools

Aperture photometry, exoplanet transit light curves, transient detection, isophote fitting and an offline minor-body catalog.

Non-destructive history

Every operation is recorded, can be toggled, reordered or re-edited, and exported as a macro or run from the Python console.

All features

Known limitations

Alpha software: bugs are expected and some tools are less refined than their PixInsight or Siril equivalents.
No camera raw support (CR2, NEF, ARW, DNG); convert to FITS or TIFF first.
GPU acceleration is NVIDIA CUDA and Apple Metal only. AMD and Intel GPUs run the CPU path. The CUDA path is verified on the maintainer's machine, not in CI.
AI Sharpen has no published model yet and falls back to Richardson-Lucy deconvolution.
First install downloads about 2 GB of PyTorch; Windows builds are unsigned, so SmartScreen may warn.

Questions

Answers to the common ones (is it free, does it need an NVIDIA card, does it upload images, is it production ready) are on the FAQ. Documentation lives in the repository: getting started, workflow, features, tools reference and the project facts every page here is checked against.