SemanticVault
Find any image by describing it — even offline.
Included in Pro
What it does
Keyword search only finds what you tagged, and nobody tags ten years of photos. SemanticVault embeds every image with an on-device CLIP model, so you can search the way you think: "sunset over the sea", "dog in the snow", "red car". No cloud, no upload, no manual labels.
Search combines with filters and boolean logic: narrow by rating, camera, lens, focal length, aperture or ISO, combine terms with AND / OR / NOT, and look only at photos with faces. Or drop in a reference image and find similar ones, draw a rough sketch, or pick a color.
Beyond search, SemanticVault clusters your library into scenes and similar groups, finds photos taken at sunrise or sunset, and gives you a raw overview grouped by camera. If the model is ever unavailable, it falls back to a classic text search instead of failing.
Facts
- On-device models
- CLIP, YOLO, FER+, OCEC, OpenCV
- Natural-language search
- Yes, 16 languages
- Boolean search
- AND / OR / NOT
- Object classes
- 80 (COCO)
Key features
- Natural-language search (CLIP)
- 512-dimension semantic embeddings
- Boolean search (AND / OR / NOT)
- Filter chips (rating, type, year, camera, focal, aperture, ISO)
- Face-only filter
- Composition filter
- Image-to-image search
- Sketch search (draw to find)
- Color search with minimum color share
- Object counting (YOLO)
- Smart collections
- Saved searches
- Similar-image clusters
- Scene clustering (k-means)
- Sunrise / sunset search
- Weather and mood chips
- RAW overview grouped by camera
- GPS map of your photos
- Favorites
- Automatic tagging
- Search history (last 50)
- Filter within results
- Search diagnostics (index and model status)
- Resumable re-indexing
- Ensemble embedding for better matches
- Gibberish guard for meaningless queries
- Queries in 16 languages
- Text-search fallback (level 2)
- Model status badge
- Scope toggle (imported or entire library)
- Collapsible advanced search