📷
Welcome to Best Photo Picker
Import a folder of photos to get started. The app will score, deduplicate, and help you pick the best ones.
1Import
2Score
3Pick
4Export
Loading…
Choose a folder of photos to add to your library. Photos are copied into the library — your originals are not moved or modified.
After import, photos are scored on sharpness, lighting, faces, and composition. This may take a few minutes for large folders.
Permanently delete all photos from disk and database.
How many photos to pick and how to spread them across your timeline.
Find and remove near-identical photos so you don't pick the same shot twice.
Uses AI to catch visually similar photos that basic hashing misses. Click below to enable.
Semantic dedup loads every photo's CLIP fingerprint into memory at once. Peak footprint is ~3x the fingerprint size: the dictionary, the search matrix, and the dedup scratch matrix all live at the same time. The default 200,000-photo cap keeps that under ~1.2 GB — comfortable on a 16 GB machine sharing memory with a browser and the OS. Enabling above the cap is safe as long as you have headroom for the reported peak; if you start seeing the app slow under memory pressure, disable below.
Basic hash matching — catches exact and near-exact copies.
Changes take effect immediately.
Controls how strict the matcher is when deciding whether two faces are the same person. Move left if strangers end up in one group; move right if one person is split across several. Regroup rebuilds all person groups from scratch using the new threshold — faces stay attached to photos, but who they're grouped with may change.
Higher finds more faces but may include false positives.
Lower catches small or distant pets but may include false positives. Default 0.20 is tuned for cats on furniture.
How many photos two people must share before they form a group in the Groups view. Lower surfaces one-off get-togethers; higher keeps only recurring constellations. Takes effect the next time Groups loads.
How confident the detector must be before recognizing a face for person grouping. Default is optimized — raise if you see non-faces (hands, toys) in person groups, lower if people are missing. Requires re-analysis.
Minimum face area as percentage of image. Lower catches distant faces in landscapes but may add noise. Default 0.20% ≈ 40×40 pixels at 1024px. Requires re-analysis.
Minimum quality score (frontality, sharpness, size) for a face to enter person grouping. Higher reduces noise but may miss profile shots. Requires re-analysis.
Higher catches smaller faces in group shots but takes longer.
Lower catches crops and slight edits too.
Larger thumbnails look sharper but use more memory.
Scan inside aliases and linked folders. Turn on if your photos are on an external drive or NAS.
Changes apply on next Analyze All.
{{ bpp_posture_statement }}
Loading…
Waiting for the server to come back.