Photos compared
—
across — AI configurations
Best AI configuration
—
F1: —
Photos with AI baseline
—
Stored in your backend database
Explicit feedback
0
✓
/
0
✗
Approved · Corrected by user
Tip: hover any control or column header for a quick description.
Read more →
| Provider | Model | Mode | Photos | Composition kept removed added | Precision | Coverage | F1 | Title kept | Caption kept | Approved | Corrected |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Loading… | |||||||||||
About these metrics
These numbers compare what the AI proposed for each photo against what you ended up keeping in your Lightroom catalog after editing. They are behavioural proxies, not ground-truth precision and recall — a removed keyword might just be your stylistic preference, an added keyword might be a personal organisational tag the AI couldn't possibly know. Trust the relative ranking between rows much more than the absolute numbers.
- Min photos hides table rows whose Photos column is below the chosen number. The default of 20 prevents drawing strong conclusions from a tiny sample. The Provider and Model dropdowns always list every configuration the backend knows about, even hidden ones, so you may see a provider in the dropdown that does not currently appear in the table — lower Min photos or tick Show all to make it visible.
- Show all bypasses the Min photos filter so every AI configuration the backend knows about appears in the table, no matter how few photos it has been used on.
- Provider and Model filter the visible table to a single AI configuration. The Model dropdown adapts to the selected Provider so you only see compatible combinations.
- Mode tells you whether the photo went through one-pass analysis (a single AI call) or two-pass (a cheap category-detection call followed by a focused tagging call).
- Composition bar shows kept (cyan) / removed (rose) / added (amber) keyword shares per row.
- Precision = kept ÷ (kept + removed) — of the AI's suggestions, how many survived.
- Coverage = kept ÷ (kept + added) — of the keywords you ended up with, how many the AI got first try.
- F1 = harmonic mean of Precision and Coverage. The row's headline score.
- Title kept — share of photos whose AI-generated title you kept unchanged.
- Caption kept — share of photos whose AI-generated caption you kept unchanged.
- Approved — photos you explicitly marked Approved using the AI Keyword Feedback field in Lightroom's Metadata panel.
- Corrected — photos you explicitly marked Corrected by user using the AI Keyword Feedback field in Lightroom's Metadata panel.
Capture a fresh snapshot from Library → Plug-in Extras → Capture Accuracy Snapshot… in Lightroom whenever you want to refresh the comparison. Keyboard shortcuts: r refresh · c download CSV.