Professional astrophotography image processing. GPU-accelerated, free, and open source alternative to PixInsight.

You don't need to install Python to use Cosmica. Download the standalone, ready-to-use versions (The AppImage is a portable CPU-only version. For full NVIDIA GPU acceleration, please install via Poetry):

• Windows: Download .exe (v0.1.9-alpha)
• Linux: Download .AppImage (v0.1.9-alpha)

(Note: This is an Alpha release. Bug reports and feedback are welcome in the Issues tab!)

Cosmica is built as a modern workflow tool from calibration to export, featuring out-of-the-box AI integration.

• Calibration — Master dark / flat / bias creation with batch light frame calibration
• Alignment — Star-based registration (1-pass, 2-pass refinement, triangle matching), FFT phase-correlation, and comet nucleus tracking
• Stacking — Sigma-clip, winsorized sigma, linear fit, percentile clip, ESD, and min/max rejection
• Drizzle Integration — 2× / 3× scale-up for undersampled data
• Multi-Session Stacking — Combine data from multiple nights with per-session adaptive weighting
• Subframe Selector — Automatic frame scoring by FWHM, eccentricity, SNR, background, and star count
• Debayer — RGGB, BGGR, GRBG, GBRG with VNG and other methods; auto-detection from FITS header

Local AI models download automatically on first use. No cloud required.

• AI Denoise — Noise2Self U-Net trained on real astro images
• AI Sharpen — Neural deconvolution for recovering fine detail
• Star Removal — StarNet-based starless image generation

• Deconvolution — Richardson-Lucy with optional total-variation regularization and deringing protection
• Spatially-Varying Deconvolution — Per-zone PSF measurement and blending for field curvature / coma correction
• Wavelet Sharpening — Multi-scale contrast enhancement
• MLT (Multiscale Linear Transform) — Selective noise reduction per frequency band
• Local Contrast Enhancement — GPU-accelerated CLAHE
• Unsharp Mask — Standard and advanced masking
• Median Filter — Impulse noise removal

• Photometric Color Calibration (PCC) — Plate solve then match against Gaia DR3
• SPCC — Spectrophotometric calibration with filter response curves
• Background Extraction — Polynomial surface fitting, ABE (RBF-based), and dynamic sample placement
• Background Neutralization — Robust color balancing from background samples
• Color Calibration — Statistical and catalog-based correction
• SCNR — Green noise reduction for narrowband and OSC images
• Color Adjustment — Saturation, hue shift, vibrance
• Curves — Per-channel curve editor with histogram overlay
• Histogram Transform — Black point, midtone, white point with live preview

• Narrowband Combine — HOO, SHO, and custom palette mappings
• LRGB Combine — Luminance-weighted RGB merging
• Channel Combine — Custom channel mapping dialog
• Continuum Subtraction — Remove broadband contamination from narrowband filters
• HDR Composition — Multi-exposure blending

• Cosmetic Correction — Hot, cold, and dead pixel repair
• Banding Reduction — Horizontal and vertical pattern removal
• Chromatic Aberration Correction — Auto-detect and manual shift
• Vignette Correction — Model-based flat-field emulation
• Star Reduction — Shrink star bloat without star removal
• Morphology — Dilate, erode, open, close for star masks
• TGV Denoise — Total generalized variation (non-AI, edge-preserving)
• PSF Measurement — Interactive FWHM, ellipticity, and angle from detected stars

• Plate Solve — ASTAP and astrometry.net with auto-fallback
• WCS Overlay — Catalog star positions drawn on the image
• DSO Annotation — Automatic deep-sky object labels from solved coordinates

• Modern Dark Theme — GitHub-inspired green accent, designed for long nights
• 4-Panel Layout — Project tree | Canvas + Histogram | Tools | Log
• Split Before/After Preview — Draggable divider with live preview on every tool
• Blink Comparator — A/B frame comparison at variable FPS
• Interactive Histogram — Log scale, per-channel stats, clip indicators
• Curve Editor — Per-channel control points with histogram backdrop
• Macro Recorder — Record and playback processing steps
• Python Console — Embedded scripting dock with live image access
• Batch Processing — Unattended folder processing
• Smart Processor — One-click automated workflow
• Equipment Profiles — Camera / telescope metadata for plate-scale calculations
• Undo / Redo — Full history stack with display-reference matching
• Presets — Save and recall tool settings

• Read: FITS, XISF, TIFF, PNG, JPEG (auto-debayer for OSC)
• Write: FITS, XISF, TIFF (8/16-bit), PNG (8/16-bit), JPEG

• Python 3.11–3.14
• Poetry (dependency management)

# Clone the repository
git clone [https://github.com/majmichu1/cosmica.git](https://github.com/majmichu1/cosmica.git)
cd cosmica

# Install dependencies
poetry install

# Run the application
poetry run cosmica
# or
poetry run python -m cosmica

poetry install --with build
poetry run pyinstaller build/cosmica.spec

poetry run pytest   # 729+ tests

poetry run ruff check .

poetry run mypy cosmica

Cosmica includes self-supervised AI training scripts. You can train your own denoise model on your astro images:

# Place your FITS files in astro_data/
mkdir -p astro_data
cp /path/to/your/*.fits astro_data/

# Train the denoise model
poetry run python scripts/train_denoise_model.py --input astro_data --epochs 30

The model uses Noise2Self — a self-supervised approach that learns to denoise from noisy images alone, without needing clean reference images.

Contributions are welcome! Please read our Contributing Guide for details on how to get started.

This project is licensed under the GNU General Public License v3.0 (GPL-3.0).

The GPL v3 is required because Cosmica uses PyQt6, which is licensed under GPL v3 for open-source use.

If you find Cosmica useful and would like to support the project:

Every contribution helps keep this project free and open source!

• PyQt6 — User interface framework
• PyTorch — GPU-accelerated computation
• Astropy — FITS file I/O
• Noise2Self — Self-supervised denoising
• All the open-source astronomical software that inspired this project

• 42 core processing modules
• 729+ tests
• 7.7M parameters in the AI denoise model
• 181k+ training patches from real astro images