The exposure values survived this small Lightroom Classic-to-darktable test. The rendered images still differed, and a red color label became purple.

Those results came from two RAW files and their genuine Adobe XMP sidecars, tested on November 13, 2026. They do not establish compatibility for a complete Lightroom catalog.

What the test actually moved

The inputs were two CC0 samples from raw.pixls.us: a Canon EOS 5D Mark IV 983.CR2 and a Sony ILCE-7M3 2414.ARW. Lightroom Classic 16.0 was used to create the edits and save metadata through its graphical interface.

Both photos received four stars, a red color label and two ordinary keywords: CC0 and fossdesk-migration. The Canon photo received +0.75 EV and a crop. The Sony photo received -0.50 EV and remained uncropped. The Adobe XMP files were generated by Lightroom, rather than written by hand to imitate its output.

In darktable 5.6.2, a fresh configuration and database were used. The RAW folder was imported with the matching Adobe <basename>.xmp files alongside the images. Each photo was then opened in darkroom to trigger the Lightroom adjustment import.

darktable wrote its own sidecars, named <basename>.<RAW-extension>.xmp. The CLI exports used those darktable-generated sidecars. Passing an Adobe sidecar directly to a CLI command would not have established that this GUI migration route worked.

The fields that survived

Both photos retained their four-star rating and both ordinary keywords. The imported exposure instances displayed +0.750 and -0.500 EV. The Canon crop produced enabled crop-related processing; the Sony photo remained uncropped.

The original RAW files and the retained Adobe sidecars had unchanged SHA-256 hashes in the recorded checks. darktable created separate processing sidecars.

These checks were deliberately narrow. Ordinary keywords are not evidence for hierarchical keyword compatibility, and a preserved exposure-module value is not evidence for identical rendering across two processing engines.

The red label became purple

The Lightroom source GUI showed red. The Adobe sidecars contained a Chinese label value and a red Photoshop label-color value. In the target configuration, both images had color-label ID 4, identified as purple in the recorded target checks.

This was a mismatch in one Chinese-source/English-target setup. Its root cause was not established. It would be misleading to call it a proven translation bug, or to claim every red label migrates incorrectly.

If color labels drive your workflow, check their meaning after a trial import rather than relying on the presence of a label field alone.

The exported dimensions differed

Lightroom reference JPEGs were exported through its GUI at quality 90 in sRGB. The migrated and no-sidecar baseline JPEGs were exported through darktable's CLI at quality 90 with an explicit sRGB profile and no requested downscaling.

Sample Lightroom reference darktable with migrated edits darktable without the Adobe sidecar
Canon 983.CR2 4912 × 3181 4929 × 3195 6744 × 4500
Sony 2414.ARW 6000 × 4000 6048 × 4024 6048 × 4024

The Canon result showed a roughly similar cropped composition, with different output dimensions. The Sony result shows why a size difference should not automatically be attributed to a failed crop import: the migrated export and the no-sidecar baseline had the same dimensions, while the Lightroom reference differed.

The recorded images also differed visibly in color and brightness. No extra matching color adjustments were applied to make them look closer. The results do not support a claim of lossless migration or a ranking of highlight-recovery quality.

An exit code was not enough

An initial darktable launch exited before adjustment import was complete. A later attempt with --disable-opencl succeeded, but the original exit's cause was not diagnosed. That does not prove a GPU fault.

One CLI export attempt also returned exit code 0 while placing output at an unexpected path when backslashes were used. The failed-attempt evidence was retained, and the corrected run used forward-slash paths. The expected outputs were then checked for existence, successful decoding, dimensions and hashes.

Those were two separate problems. Neither should be hidden behind a single green “migration passed” label.

What this means for a trial migration

Make copies of a few photos that use the metadata and adjustments you depend on. Preserve the source RAW files and Adobe sidecars, and record what you changed before importing. Check stars, keyword contents, label meanings, adjustment values and the resulting image separately.

Export reference images from both applications, inspect their dimensions and appearance, and reload the target application before trusting the result. Keep failures alongside successful outputs.

This test did not import a .lrcat catalog or cover collections, virtual copies, full edit history, presets, AI masks, GPS, hierarchical keywords, other operating systems or other versions.

The test was recorded for FOSS Desk, which I maintain. The Lightroom-to-darktable guide includes the scoped results and limitations. This is an independent report, unaffiliated with Adobe or the listed projects.

AI assistance was used to organize and draft this article from recorded project evidence. It does not add new migration experiments or establish causes that were not measured.