The new version of the open-source editor darktable 5.6 integrates artificial intelligence capabilities that run locally on the user's device via ONNX models.
What Happened
Darktable developers have introduced version 5.6, which includes AI Object Mask tools for one-click automatic creation of vector object masks, as well as the Neural Restore module. The latter provides noise reduction (for both RAW and post-processed files) and upscaling using specialized RealPLKSR and BSRGAN models. All computations are performed directly on the user's hardware without contacting cloud services.
Context
The integration of local ONNX models represents a shift toward decentralized and private AI usage in professional software. This allows for the use of modern SOTA architectures for heavy tasks, such as segmentation and image restoration, while maintaining complete application autonomy.
Why It Matters for the Industry
This is an important precedent for the open-source industry, demonstrating a viable model for AI implementation that does not depend on closed APIs or cloud infrastructure. Such an approach could slow the dominance of cloud SaaS solutions in the creative tools sector, setting a standard for local AI inference in desktop professional software.
Why It Matters for Users
Professionals gain access to advanced automation features for routine tasks, such as object selection and image quality enhancement, without the need for paid subscriptions or a constant internet connection. This significantly accelerates the processing workflow and guarantees complete data privacy.
Sources
Author
Look at AI, Editorial Team