[Feature] Automatic "Best Photo" / Great Shot Detection

[Matrinica]

I have searched the existing feature requests, both open and closed, to make sure this is not a duplicate request.

• Yes

The feature

[Feature] Automatic "Best Photo" / Great Shot Detection

Is your feature request related to a problem? Please describe.
As my photo library grows into the tens of thousands, it becomes increasingly difficult to surface the truly great shots — the photos worth printing, sharing, or revisiting. Currently, finding them requires either manually reviewing everything or relying on CLIP keyword searches ("beautiful sunset"), which doesn't capture compositional quality or personal preference.

I'm looking for a way to automatically surface great photos based on two signals:

1. Photography quality guidelines — composition (rule of thirds, leading lines, symmetry), exposure, focus/sharpness, color harmony, subject salience, lack of blur/clutter
2. Personal preference — photos I've already marked as favorites or rated highly with stars

Describe the solution you'd like
A dual-mode "Great Shots" feature:

• AI-powered quality scoring — A machine learning model (either built-in or as a plug-in ML model for immich-ml) that analyzes photos and assigns an aesthetic/composition quality score. High-scoring photos are surfaced in a dedicated view or smart album. This could draw on proven approaches like NIMA, aesthetic quality assessment models, or composition-aware CNNs.

• Personal preference boost — Photos you've favorited or given high star ratings (4-5 stars) are automatically included or ranked higher in the same view, so the "best photos" view combines objective quality + your personal taste.

• Dedicated view — Something like a "Best of [Year]" smart album or timeline view that surfaces these candidates. Bonus if it can auto-generate a yearly highlight reel.

• Blurry/low-quality filtering (as a complement) — A maintenance/scanner job that flags blurry, underexposed, or otherwise technically poor photos (similar to the already-requested "blurry photo detection" feature).

Describe alternatives you've considered

• Manual favorites + star ratings: works but doesn't scale for large libraries
• CLIP contextual search ("beautiful photo"): semantic, not compositional — misses technically good photos of mundane subjects
• External pipelines: running a separate ML model and tagging via API, but this is complex and not integrated
• Third-party tools (immich-power-tools, etc.): community projects come and go; would prefer core feature

Additional context
Existing infrastructure that could support this:

• CLIP-based smart search and the VectorChord vector database already process images — an aesthetic scoring model could integrate similarly
• Star ratings (recently shipped on mobile) and favorites provide the personal preference signal
• The machine learning pipeline is already extensible via immich-ml

Would be incredible to see this as an optional immich-ml model that generates a "quality_score" or "aesthetic_score" metadata field on assets, making it searchable and filterable.

Platform

• Server
• Web
• Mobile

[bo0tzz]

#7202