# Darktable Releases 5.6 With Optional AI Tools > Source: > Published: 2026-06-21 17:05:40.844573+00:00 # Darktable Releases 5.6 With Optional AI Tools The open-source photo editor **darktable** released **version 5.6.0**, announced on darktable.org on June 21, 2026. The release adds an optional AI subsystem, buildable with -DUSE_AI=ON, that includes object masking reported to use SAM2.1 and SegNext and a neural restore module, according to the project's release notes and LinuxCompatible. The AI features are disabled by default, per reporting by LinuxCompatible, and require additional runtime setup on Linux for GPU inference. The release also delivers performance fixes listed on darktable.org, including OpenCL optimizations, doubled darkroom preview resolution, and faster thumbnail cache builds. The GitHub release and release notes provide checksums for binaries and the source tarball. Coverage appears in Tux Machines and LinuxCompatible, which highlight UI improvements, new Lua AI inference functions, and several bug fixes. ### What happened **darktable** released **version 5.6.0**, published on darktable.org on June 21, 2026, with the project providing checksums and a downloadable darktable-5.6.0.tar.xz source archive and platform binaries, per the project's release post and the GitHub release notes. The release notes list **1561 commits**, **705 pull requests handled**, and **63 issues closed** since the **5.4** series, figures included in the darktable.org announcement. The release introduces an optional AI subsystem (build with -DUSE_AI=ON) and bundles several non-AI performance and workflow improvements, as described in the official release notes and on the darktable project site. ### Technical details (reported) The optional AI subsystem, documented by darktable and covered by LinuxCompatible, includes an **object mask** tool; LinuxCompatible reports it uses SAM2.1 or SegNext models for automated selection, and a **neural restore** module for content-aware restoration. LinuxCompatible reports the AI modules remain disabled by default so older systems stay responsive. The project also added Lua scripting functions for AI inference, and the release notes list OpenCL processing optimizations, a doubling of darkroom preview resolution, and a change to use embedded JPEGs for thumbnail generation to speed cache builds. ### Editorial analysis - technical context Tools that offer optional, off-by-default AI subsystems reflect a broader pattern in desktop imaging software, where maintainers balance adding model-driven features against resource, dependency, and UX costs. For practitioners, given reports of SAM2.1 and SegNext as supported backends, this signals reliance on established segmentation models rather than bespoke, monolithic networks. Industry-pattern observations: enabling GPU inference across Linux, Windows, and macOS typically requires packaging, runtime drivers, and distribution-specific install scripts, so keeping AI optional reduces friction for conservative users while allowing power users to enable model-based workflows. ### Context and significance This release is notable because it brings model-driven editing features to a mature open-source RAW pipeline, while preserving non-destructive workflows. For photographers and imaging engineers, the combination of faster core processing, UI refinements, and scriptable AI inference points to a practical path for integrating vision models into batch and programmatic editing. The release also matters to downstream packagers and distributions because the project cautions about building from the correct source tarball and notes that library and configuration changes between **5.4** and **5.6** will not be backward compatible. ### What to watch Observers should watch adoption notes from distributions and package maintainers for prebuilt AI-enabled packages, the documentation contributions the project requested for completing release docs, and community feedback on model performance and accuracy in diverse RAW datasets. Also track whether GPU runtime install scripts for Linux are adopted widely, and how the Lua AI inference hooks are used in third-party automation or batch processing scripts. ## Scoring Rationale The release brings model-driven features to a widely used open-source RAW editor and adds scriptable AI hooks, which is notable for practitioners integrating vision models into production or automated photo workflows. It is important but not industry-shaking. Practice interview problems based on real data 1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with. [Try 250 free problems](/problems)