In last month's issue of Linux Magazine [1], I wrote about the darktable interface, its general philosophy, and the first steps required to import and organize images. Once files are safely in the library, the real strength of darktable emerges in the darkroom view, where raw sensor data is transformed into a visually meaningful photograph. This transformation is not simply a matter of increasing contrast or saturation, but the result of a carefully designed processing pipeline grounded in color science, perceptual modeling, and a strict separation between scene data and display constraints.

With version 5.4, darktable firmly promotes a scene-referred workflow, a paradigm that differs significantly from the display-referred approach that many photographers have internalized through years of using older raw processors or commercial software. Understanding this workflow is essential to achieving consistent, high-quality results and to avoiding frustration when familiar controls appear to behave in unexpected ways. Modules such as filmic rgb and sigmoid are not isolated tools, but key elements of a coherent pipeline whose logic becomes clear only when the scene-referred philosophy is fully understood.

Scene-Referred vs. Display-Referred

The fundamental difference between scene-referred and display-referred editing lies in the reference point used during image processing. In a display-referred workflow, the image is adjusted as it appears on a specific output device, usually a monitor calibrated to sRGB or a similar standard. Brightness, contrast, and color are manipulated directly on data that has already been tone mapped and gamma encoded. This approach often feels intuitive because every adjustment produces an immediate visual response, but it merges two distinct stages of the imaging process: the representation of the original scene and its adaptation to the limitations of a display. As a result, edits that look correct on one screen may not translate well to another, and exporting the same image for print or web frequently requires additional, image-specific corrections.

[...]