darktable: Non-Destructive RAW Photo Editor and Digital Darkroom

Non-Destructive RAW Development for Photographers

darktable is designed for photographers who work with RAW files and need a non-destructive editing workflow. All edits are stored as metadata in a SQLite database rather than modifying the original file. The original RAW image is never altered. This means every editing decision can be reversed at any point, and multiple editing histories can exist for the same source file.

The application organizes work into two main views. The lighttable is a zoomable browser for managing and rating the image collection. The darkroom is the editing environment where individual images are processed through a module pipeline. Both are described in the README as a virtual lighttable and darkroom.

darktable explicitly states in its README that it is not a free Adobe Lightroom replacement. The README uses bold text to make this clear. The workflows and module organization are different enough that users transitioning from Lightroom will need to learn darktable's specific approach rather than expecting a feature-for-feature correspondence.

Platform Support and Hardware Requirements

darktable runs on Linux, FreeBSD, NetBSD, OpenBSD, Windows 10 and later, and Apple Silicon Macs running macOS 14 or later. The README notes that Linux is the primary development platform and that other platforms may have additional bugs not present in the Linux version. Windows specifically lacks printing support.

The hardware requirements from the README:

- RAM: 4 GB minimum, 8 GB recommended - CPU: Intel Pentium 4 minimum (Core 2 for Windows), Intel Core i5 at 4x2.4 GHz recommended - GPU: optional but recommended, Nvidia with 1024 CUDA cores and 4 GB, OpenCL 1.2 compatible - Free disk space: 250 MB minimum, 1 GB recommended

The README notes that darktable can run on lightweight configurations including a Raspberry Pi, but that modules like denoise, local contrast, contrast equalizer, retouch, and liquify will be too slow for practical use on low-end hardware. A GPU is strongly recommended for a smoother experience, and Nvidia GPUs are specifically recommended over AMD due to some AMD drivers behaving unreliably with certain modules.

Big-endian platforms are not supported. 32-bit platforms are not officially supported, though they may work.

Installing darktable

Precompiled releases for Linux, Windows, and macOS are available at darktable.org. The README links to the releases page for current stable and development snapshot downloads.

For Linux, the top-level `build.sh` script in the repository handles compilation with CMake. The script accepts `--enable-ai` to compile with AI feature support, which corresponds to the CMake option `-DUSE_AI=ON`. AI features are disabled by default in the build system. When enabled, they are also disabled by default in preferences and must be activated by the user. Models for AI inference are downloaded separately from the AI tab in preferences, with the default model repository at darktable-org/darktable-ai.

The current stable release is 5.6.1 (2026-08-27). Nightly builds are available for users who want the latest development state. The last push to the master branch was on 2026-09-27.

AI Features: Optional Denoise, Upscale, and Object Masks

darktable includes optional AI-powered modules for denoise, upscaling, and object masks. These require building with `-DUSE_AI=ON` or the `--enable-ai` build flag, and enabling the features in preferences after installation.

CPU inference is bundled and works without additional software. GPU acceleration requires installing a GPU-enabled ONNX Runtime separately:

curl -fsSL https://raw.githubusercontent.com/darktable-org/darktable/HEAD/tools/ai/install

For GPU support, the requirements differ by GPU vendor. Nvidia requires CUDA Toolkit 12 or 13 and cuDNN 9.x. AMD requires ROCm 6 or later and MIGraphX on Linux only. Intel requires OpenCL or Level Zero drivers plus OpenVINO Toolkit on Windows. Windows DirectML acceleration is bundled and requires no extra install on Windows 10 1903 or later. macOS CoreML is bundled for Apple Silicon.

The README warns that on AMD GPUs, the first inference run per model is significantly slower because ROCm compiles the model graph on the fly. The compiled graph is cached, so subsequent runs are faster. Teams evaluating darktable for AI-assisted editing on AMD hardware should account for this first-run latency during testing.

Lua Scripting and Extension Points

darktable includes a Lua scripting API that allows automating tasks and extending the application's functionality. The Lua API documentation is maintained in a separate repository: darktable-org/luadocs. User scripts can define custom export pipelines, integrate with external tools, and add menu items.

The website and user manual are also maintained in separate repositories (dtdocs and dtorg), which reflects the project's approach to keeping documentation, website, and application source in separate version control streams. Translators working on the user interface contribute to the `po/` directory in the main repository.

The repository uses CMake with a `ConfigureChecks.cmake` and `DefineOptions.cmake` file for build configuration. Feature flags like `USE_AI` are defined in `DefineOptions.cmake`. Developers building custom configurations or packaging darktable for a distribution should review both files to understand the available options.

Limitations and Maintenance Considerations

darktable's module pipeline is deep and takes significant time to learn. The darkroom has dozens of modules with controls that interact with each other, and the expected workflow differs from Lightroom's adjustment panels. Users who switch from other editors should expect a substantial learning period before reaching proficiency.

The Windows version lacks printing support. Users who need to print directly from darktable should use the Linux version, or export images to another application for printing on Windows.

The AI features have a hard requirement for ONNX Runtime for GPU acceleration, and GPU memory requirements of at least 4 GB VRAM. If GPU inference fails due to out-of-memory or an unsupported operation, darktable falls back to CPU inference automatically, but CPU inference is substantially slower.

The GPL-3.0 license requires that any distributed modification to darktable's source code be released under the same license. This affects commercial software vendors who might want to embed or extend darktable but do not want to release their modifications publicly.

darktable Against Adobe Lightroom

Adobe Lightroom Classic is a commercial RAW workflow application sold as part of an Adobe Creative Cloud subscription. It runs on Windows and macOS and provides a catalog-based workflow, adjustment panels, masking tools, and cloud sync features.

The README specifically states darktable is not a Lightroom replacement, which is a direct acknowledgement that the two applications have different philosophies. Lightroom organizes work around a linear adjustment panel with sliders that map to photographic concepts like exposure, contrast, and highlights. darktable organizes work around a module pipeline that exposes the underlying signal processing operations more explicitly.

For photographers who want free software on Linux, darktable is the primary option in its category. For photographers on Windows or macOS who are already familiar with Lightroom's workflow, the transition to darktable requires unlearning and relearning how adjustments are applied.

RawTherapee is another open-source RAW editor in the same category. It shares many platform targets with darktable but has its own module organization and defaults. The README does not discuss RawTherapee.

Editorial conclusion

darktable is the right tool for photographers who want a free, open-source, non-destructive RAW workflow on Linux, Windows, or Apple Silicon, and who are willing to invest time learning its darkroom module system. It is not positioned as a Lightroom replacement: the README states this explicitly. Before adopting it, verify that your camera is supported by checking the project's camera support documentation, and confirm whether the AI features you want (denoise, upscaling, object masks) require the optional `-DUSE_AI=ON` build flag, which is not enabled in all package manager distributions.

Frequently asked questions

Official sources

1. darktable-org/darktable on GitHub
2. License: GPL-3.0
3. Project website
4. README
5. Releases