Duecki1/CalibRaw

What it is

CalibRaw is a free, open-source RAW photo editor for Linux, Windows, macOS and Android, licensed under GPL-3.0-or-later. It combines a scene-referred development pipeline (Light, Tone Curve, Color, Color Mixer, Color Grading, Detail, Optics, Effects, Masks) with local AI tools: subject/sky/depth/object masks, AI Remove and AI Denoise. All processing is GPU-based through Vulkan, Direct3D 12 or Metal.

RAW decoding relies on LibRaw plus Rawler for DNG and JPEG-XL DNG streams, and the same engine and .calibraw sidecars are shared between the desktop and Android apps. Edits are non-destructive; presets, .cube LUTs, HDR merge, focus stacking and a headless batch export CLI are included.

Quick start

1. Download the build for your platform from the latest release: .AppImage/.deb or Flathub on Linux, .apk or F-Droid on Android, installer or portable .zip on Windows, .zip on macOS.
2. Check the graphics backend before starting. On Linux, a Vulkan driver is required; machines without one can fall back to Mesa's software driver lavapipe (slow). You can force an API with WGPU_BACKEND, for example WGPU_BACKEND=vulkan.
3. Open a folder in the Library to browse thumbnails, then open a photo in Develop to adjust sliders, curves and masks.
4. AI tools download their models once, after you agree; models are SHA-256 verified and then run offline.

For batch rendering from the terminal, the calibraw-develop-export CLI is available:

calibraw-develop-export --input photo.ARW --output photo.jxl

Building from source needs Rust 1.92.0+, libclang, cmake, pkg-config and standard C/C++ build tools:

git clone https://github.com/Duecki1/CalibRaw.git
cd CalibRaw
cargo run -p calibraw-ui --bin calibraw --release

Requirements and constraints

• GPU rendering via Vulkan, Direct3D 12 or Metal is mandatory; OpenGL-only drivers are unsupported. Windows falls back to the WARP software renderer, macOS requires Metal.
• Linux .deb packages target Debian 13+ and Ubuntu 24.04+.
• AI models are downloaded on first use, so an initial network connection and disk space are needed; afterwards inference is offline. Editing and AI tools do not send photo data off-device.
• Camera support is bounded by LibRaw and Rawler; check their maintained camera lists for specific bodies.
• Imported .xmp presets are translated on load, with unsupported settings preserved but not applied.

HotRepo's take

On the Rust daily list for 2026-11-02, CalibRaw recorded 222 stars and 14 forks, gaining 19 stars that day and reaching position 11 on the Rust chart. That is a single snapshot, so it shows the project attracted attention on a Rust-language list, not a broad all-language trend. The distinguishing technical claim is parity between desktop and Android on one Rust/wgpu engine with fully local AI models — a combination that separates it from darktable's module-level workflow and RapidRAW's interface inspiration noted in the README.