StarNet Software

AI-powered software for star removal and noise reduction in astrophotography.

Command-line Tools

This page describes the standalone StarNet and DeepSNR command-line packages. Use these files for command-line workflows, automation, integration with other image processing software, and standalone processing outside PixInsight.

Command-line tool releases

Command-line archives are for standalone processing, automation, and applications that call StarNet or DeepSNR executables directly. For example, Siril's StarNet integration uses the command-line StarNet installation path.

Latest version feed

Building an application or script that uses StarNet2 or DeepSNR? Our public JSON feed lets your software check the latest CLI versions for each platform, notify users when an update is available and get direct archive and installer download links. Learn how to use the latest version feed.

Release matrix

Current command-line packages are published for Linux x64, Windows x64, macOS x64, and macOS ARM64. StarNet2 and DeepSNR follow the same targets as the PixInsight modules, but command-line releases can move on their own schedule and are packaged as separate files for terminal use, scripts, and host applications that call external executables.

Current ORT/CoreML command-line packages are self-contained. Users do not install a separate TensorFlow, Torch, or backend-runtime package.

Current StarNet2 command-line release matrix

Operating system Architecture Version Backend Release date
Linux x64 2.6.2 ONNX/ORT 2026-09-19
Windows x64 2.6.2 ONNX/ORT + DirectML 2026-09-19
macOS x64 2.6.2 ONNX/ORT 2026-09-19
macOS ARM64 2.6.2 CoreML 2026-09-19

Current DeepSNR command-line release matrix

Operating system Architecture Version Backend Release date
Linux x64 1.3.2 ONNX/ORT 2026-09-19
Windows x64 1.3.2 ONNX/ORT + DirectML 2026-09-19
macOS x64 1.3.2 ONNX/ORT 2026-09-19
macOS ARM64 1.3.2 CoreML 2026-09-19

System and build information

Current command-line packages are self-contained. The table below separates what users need to run the package from the systems used to build and validate the release.

Current command-line system information

Target Runtime requirement Built and validated on
Linux x64 x86_64 Linux on a compatible glibc-based Linux distribution. The portable archive works across supported distributions; the .sh installer is for Debian-compatible systems. Ubuntu 24.04.2 / WSL2 with bundled ONNX/ORT CPU.
Windows x64 64-bit Windows 10 or 11. Windows 11 x64.
macOS x64 macOS 13.4 or newer. Intel Mac, or Apple Silicon Mac through Rosetta 2. macOS 26.5 on Apple Silicon Mac through Rosetta 2.
macOS ARM64 Apple Silicon Mac with macOS 13.1 or newer. macOS 26.5 Tahoe on Apple Silicon Mac.

Current command-line packages include their required runtime files. They do not require separate TensorFlow, Torch, CUDA, Python, Visual Studio, ONNX Runtime, or CoreML runtime installs.

Release notes

StarNet2 2.6.2 and DeepSNR 1.3.2 improve linear FITS handling, output fidelity, and processing diagnostics.

StarNet2

  • Improved intermediate 2× processing and consistency of difference and unscreen star layers.
  • Corrected integer output encoding and grayscale processing, and avoided unnecessary neural processing of wholly constant image tiles.

DeepSNR

  • Corrected integer output encoding and grayscale processing, and avoided unnecessary neural processing of wholly constant image tiles.
  • Improved model lookup and backend fallback diagnostics. Model 2 supports RGB and grayscale; model 1 requires RGB with non-identical planes, including with custom weights.

Shared command-line behavior

  • Linear FITS input can contain negative calibration samples. Values above the supported white level are clipped with a warning; float ADU data needs upstream normalization.
  • Improved output precision and input-range diagnostics. --quiet suppresses routine messages while retaining warnings and errors.
  • See the StarNet and DeepSNR references for input scaling, output units, and clipping behavior.

Previous StarNet2 releases

  • 2.6.0 / 2.6.1: Added direct linear-image processing, compressed and normalized float FITS input, preserved astronomy metadata and source FITS representation, and automatic TIFF/PNG depth selection. The Linux patch improved startup portability.
  • 2.5.4: Added machine-readable CLI metadata and progress output for wrapper integrations, plus standard FITS input and float32 FITS output.
  • 2.5.3: Added the highlight-protection option and DirectML acceleration with CPU fallback for Windows ONNX Runtime packages.
  • 2.5.0: Added --unscreen star-layer output and native CoreML packages for Apple Silicon Macs.

Previous DeepSNR releases

  • 1.3.0 / 1.3.1: Added direct linear-image processing, compressed and normalized float FITS input, preserved astronomy metadata and source FITS representation, and automatic TIFF/PNG depth selection. The Linux patch improved startup portability.
  • 1.2.3: Added machine-readable CLI metadata and progress output for wrapper integrations, plus standard FITS input and float32 FITS output.
  • 1.2.2: Added DirectML acceleration with CPU fallback to the Windows ONNX Runtime packages.
  • 1.2.0: Added native CoreML packages for Apple Silicon Macs.

Command-line download pages