
Before troubleshooting an empty first face scan in digiKam, verify that its required deep-learning models have been downloaded. Missing model data is a prerequisite problem; changing recognition thresholds or entering more names cannot substitute for that dependency.
Check the dependency before scanning the archive
The People View manual says the models may be downloaded during first run or through Settings → Miscellaneous → System → Download required binary data.
Open that route and inspect the state actually shown by your installed version. If downloading fails, retain the reported reason and resolve the application's access to the required files. Do not assume that an ordinary webpage loading in a browser proves the application's own download completed.
This guide does not claim to have downloaded a model or measured scan performance. It also does not establish that missing data explains every failed scan.
Test detection before recognition
Once the required data is available, select a small authorized album under People rather than immediately starting a full collection scan. In the settings, choose a detection-appropriate scan option for new images.
Recognize faces only is intended for faces already detected. It is not a replacement for the first detection pass. If the photograph has no face region yet, a recognition-only attempt does not test whether the model can find a face.
Use the Search-in scope to identify the intended album. Note which sample is being processed so a scan of an empty or unrelated selection is not mistaken for a model failure.
Classify the first result
After the proposed scan, distinguish these outcomes:
- No task starts or the application still reports a dependency issue.
- The task runs but creates no face regions for the chosen sample.
- Regions appear under Unknown but no person's name has been assigned.
The third outcome is not the same as failure to detect a face. Naming and recognition require their own training and confirmation workflow.
Use the documented face-region confirmation route when proceeding. An ordinary whole-image keyword is a different data object; review catalog metadata authority before assuming that a label shown elsewhere supplies the required face evidence.
Keep the diagnosis bounded
For a sample with no regions, record the selected scan option, dependency state, image availability and the visible outcome. A single difficult photograph does not establish a general failure rate. Avoid sweeping resets that discard useful existing work before narrowing the cause.
After detection succeeds, inspect suggested identities before organizing person groups. A mistaken People grouping is a later identity-review problem; it cannot be solved by downloading the model again. The linked example uses Apple Photos, while the prerequisite controls here belong to digiKam.
This task is complete when prerequisites are verified and a limited detection test has an interpretable outcome. Accurate recognition, consent to identify people and release of the resulting face metadata remain separate decisions.
Related tasks
- Lightroom Metadata Conflict: Keep the Disk or Catalog Version?
- Should I Merge Two Photos People Groups or Correct Their Names?
Sources and scope
- digiKam Manual: People View: Face detection, recognition, model availability and confirmation are distinct stages. Assigning an ordinary tag does not create the same training evidence as confirming a face.
Documentation-based guidance researched on 23 October 2026 (Asia/Shanghai), with proposed checks rather than reported application tests. RawPedia and digiKam manual pages were opened or returned readable official body text. ExifTool official pages were readable through public search; direct PDF/XMP tag-reference fetches returned HTTP 403. Verify installed-version controls, metadata representations and receiving-reader behavior. Decorative cover assets do not depict software or a test result.
Prepared by Imagild Editorial, the company editorial team responsible for these guides. AI assisted research and drafting. This is documentation-based guidance with proposed examples, not a claim of hands-on testing or an independent product review. Report a correction.
Cover: Mak Cézar / Pexels, Pexels License. This is an illustrative source photograph, not an application screenshot or a generated test result.