Skip to main content

AI Label Generation

With a local model activated, Alcedo Studio can generate content labels for images. Labels are inferred by the on-device vision model, describe the main content of the image, and also back semantic search and the inspector's By Labels filter.

This page covers the local labeling flow. Descriptions, ratings, and rating reasons produced by an API provider are a separate flow; see API provider configuration.

The import-time prompt​

When importing images, if the project has an activated local model and the import preference is Always Ask (the default), a Use AI to analyze image content? dialog appears:

  • Generate: generates labels for this batch. The dialog then switches to Generating semantic labels, showing a circular progress, processed count, elapsed time, and estimated time remaining; you can click Cancel while processing.
  • Skip: skips this batch; the images import normally without labels.
  • Remember My Choice: remembers the current choice and syncs it into the import preference in the local AI settings.

If the import preference is Always, generation runs directly after import; with Always Skip, nothing is generated and no dialog appears. The preference can be changed in Settings → Local Content Recognition → Import.

If the project has no activated local model, importing shows Set up an AI model to analyze images; click Set up model to go to the local AI settings, or Not now to skip this batch only.

Manual generation and regeneration​

In the AI content recognition section of Local Content Recognition:

  • The top shows the current album's Images, With labels, and Need labels counts.
  • Generate only creates labels for images that do not have them yet.
  • Regenerate ignores existing labels and regenerates for all images in the album — useful after switching local models to re-label everything consistently.
  • While running, the button becomes Cancel, with progress and estimated time remaining below.

Viewing labels​

Generated labels can be viewed in these places:

  • Below each thumbnail in the grid / list.
  • Right-click an image and choose Details — in the Primary, Related, and Also Matched sections of the details dialog.
  • The By Labels area of the right inspector — click a label to filter images.

Notes​

  • Label quality depends on the local model; different model variants may produce different label sets.
  • After switching local models, use Regenerate to update the labels consistently.
  • Label generation requires the local model to be Active; an API provider's API key does not make the local labeling flow available by itself.
  • Generation speed depends on CPU, GPU, memory, image count, and model size.