AI photo tagging

Goal: understand how FrameCourier uses AI to automatically analyse your photos and populate searchable metadata fields.

How it works

When a photo is uploaded to a gallery, a background job sends a downscaled copy of the image to an AI vision model. The model reads the image and populates a set of metadata fields on the photo record. This happens automatically — you don't need to trigger it.

The following fields are populated for each analysed photo:

  • Description: a one-sentence natural-language description of the photo.
  • Tags: comma-separated keyword tags (e.g. "outdoor, sunset, family, beach").
  • OCR text: any legible text visible in the image (signs, banners, receipts).
  • Scene type: portrait, landscape, event, food, architecture, etc.
  • Setting: indoor, outdoor, studio, beach, church, etc.
  • Category: high-level classification such as wedding, portrait, family, corporate, or landscape.
  • Dominant color: the most prominent colour in the image.
  • Mood: cheerful, dramatic, serene, formal, etc.
These AI-populated fields are indexed for full-text search, meaning they contribute to search results when you use the unified search. You don't need to do anything extra — once a photo is analysed, it becomes searchable by its AI metadata.

Timing and priority

AI tagging runs as a low-priority background task. This means it is processed after higher-priority jobs like thumbnail generation, watermarking, and EXIF embedding. For a typical upload your thumbnails appear first, and the AI metadata follows shortly after.

Because AI analysis takes longer than building a thumbnail, it runs separately as a low-priority task so it doesn't slow down the rest of your uploads.

AI is built in

AI photo tagging is included with FrameCourier and runs automatically — there's nothing to configure. Each AI feature uses our managed AI service:

  • Photo tagging — a vision model analyses each image.
  • Video summary — a text model summarises the transcript.
  • Video repurposing — a text model identifies clip segments.
  • Video transcription — speech-to-text generates subtitles.

If an AI job fails (for example, a temporary service error), the system retries automatically before marking it as failed.

Viewing AI-tagged photos

The AI metadata is surfaced through search results — when you search for a keyword that matches an AI tag or description, matching photos appear in the results.

Photos without AI metadata

Photos are analysed automatically when they're uploaded, so new uploads always receive AI metadata. If you notice a photo that doesn't have it yet, just wait a moment — the analysis runs in the background and usually finishes within a minute or two of the upload.

Quality & focus review

The same analysis pass also scores each photo for sharpness and overall technical quality. On a gallery's photo management page, if a handful of photos come out noticeably softer or lower-quality than the rest of that gallery, a banner appears — "Might be worth a second look" — with a Review them button that filters the grid down to just those photos, so you can catch a soft or out-of-focus shot before your client sees it.

  • The comparison is always relative to that gallery's own photos, not some fixed bar — a consistently moody or soft-focus style won't get flagged just because something has to be at the bottom.
  • Intentional shallow depth of field (a sharp subject with a blurred background) is not treated as a defect.
  • The banner only appears once a gallery has enough analysed photos to make a fair comparison, and only when both the sharpness and quality signals agree a photo is an outlier.
  • Dismissing the banner (×) hides it for that gallery, but it reappears automatically if a newly uploaded photo also gets flagged.

Related