You rarely remember the date or filename of a photo — you remember what was in it. Description search lets you find photos the way your memory actually works: "sunset at the beach", "red bicycle", "grandma's birthday cake". And with local AI, that search runs on your own computer, so your photos are never uploaded for analysis.
Traditional search matches filenames and metadata. Description search matches the content of the image itself — what's actually in the frame. Type what you remember seeing, and the search finds the matching photos.
Cloud-based image search uploads your photos to a server for analysis. Local AI search runs the model on your own hardware (via ONNX Runtime), so nothing leaves your machine — the privacy is the point.
Description search is most powerful combined with filters. Narrow to a date range or folder first, then search within the results — this is faster and more precise than searching the whole library at once.
When you've found one good photo, similarity search finds other photos that look like it — different angles of the same moment, or shots from the same shoot.
Local AI search isn't limited to English — Lap supports search in 50+ languages, so you can describe photos in whatever language you think in.
| Method | Finds | Requires upload |
|---|---|---|
| Filename search | Text in names | No |
| Metadata search | Date/camera/location | No |
| Cloud AI search | Image content | Yes |
| Local AI search | Image content | No |
Lap's local AI search (CLIP + ONNX Runtime) finds photos by text description, visual similarity and detected subjects — running entirely on your machine, offline, in 50+ languages.
Search-by-description used to mean the cloud: upload everything, let a server model tag it, and query the result. The local version inverts that. A small model runs on your own machine, writes a plain-language description next to each photo, and your queries match against that local index. Same magic, different owner.
Practically, you install the app, point it at your folders, and walk away while it describes the archive in the background. Later, typing "two kids building a sandcastle" into Lap's photo viewer returns exactly those shots — including photos you took years before installing anything, because the model works retroactively on your existing files.
Accuracy has honest limits: it finds scenes and subjects reliably, but won't identify specific people the way face recognition does (that's a separate, deliberate feature). For "what did we photograph that trip", though, it is startlingly good — and the descriptions are editable when you want to teach it your vocabulary.
Description search doesn't replace organization; it removes the pressure of perfect organization. Folders answer "when and where was this trip"; search answers "which photos have a red kayak". You need both, because some queries are spatial and some are semantic, and no filing scheme captures semantics.
The efficient workflow: keep a loose year/event tree, invest your energy in triage instead of taxonomy, and let search do retrieval. In practice that means you browse by folder in Lap when you know the occasion, and search when you only know the content. Both hit the same local index, so neither path is slower.
One habit amplifies everything: after a shoot, one pass of culling plus one descriptive query a day keeps the mental map fresh. Weeks later, "the dinner with the long table" resolves in seconds. Your memory stops being the only index of your photo library — the library finally answers to words.
No — the AI models run locally via ONNX Runtime, so search works fully offline.
No. All AI processing happens on your machine — your photos never leave your computer.
Lap supports search in 50+ languages.
Lap