Feature ยท AI Find

AI photo search, in your own words

Somewhere on your Mac is a photo of a red umbrella in the rain, and you know it exists. Filenames like IMG_4832.HEIC give you nothing to search for, so you scroll thumbnails folder by folder. PhotoCore reads the content of every image on device and finds that photo from a plain description, in under 50ms across 10,000 photos.

Search by meaningUnder 50ms100+ languagesOn device
The problem

Why filename search fails

Finder and Spotlight match file names and a little metadata. A photo you never renamed carries no words to match, so the umbrella, the rainy street and the person under it are invisible to them. You end up scanning thumbnails by eye, and a two-minute task turns into twenty. PhotoCore takes another route: it looks at the pixels, so what you type is matched against what each image actually shows.

How it works

Describe the photo, get the photo

A model that reads images, on your Mac

PhotoCore runs SigLIP 2 through Core ML on the Apple Neural Engine. During indexing, each photo becomes a 768-dimension embedding stored in a local SQLite database. When you search, your words become an embedding too, and cosine similarity via Accelerate.vDSP ranks every image against them. The result: a search across 10,000 photos answers in under 50ms, in the language you think in.

  • Type 'red umbrella in the rain' and press return
  • Results in under 50ms across 10,000 photos
  • Search in more than 100 languages
PhotoCore
Similar shots

Search with an image instead of words

Sometimes the best query is another photo. Pick any image in your library and PhotoCore finds the shots that look like it. You can also drag and drop a reference image that was never indexed, straight from an email or a website, and search your archive against it. A tolerance slider decides how close a match must be, from near-duplicates to loosely related scenes, in visual or hybrid mode.

Privacy

Private by design

Every step runs on your Mac. There is no cloud, no account and no tracker, and search keeps working offline. Folders are indexed in place and read-only, including folders on external drives and NAS shares, which stay searchable after you disconnect them. Curious about the internals? Read how semantic photo search works, or see the everyday version on find photos by description. AI Find is one of seven tools on the features page.

FAQ

Frequently asked questions

Does AI photo search work offline?
Yes. The SigLIP 2 model runs on device through Core ML, so search works with no internet connection. It also keeps finding photos on external drives and NAS shares after you unplug them, because the embeddings live in a local SQLite database.
What can I type in the search box?
Plain descriptions of what the image shows, like 'red umbrella in the rain' or 'handwritten diagram on a whiteboard'. You can search in more than 100 languages and add filters for people, people count, quality, place and date.
How fast is search on a big library?
A search across 10,000 photos answers in under 50ms. Embeddings are computed once, during indexing, so every search after that is a local lookup ranked with cosine similarity.
Which image formats does PhotoCore index?
JPEG, PNG, HEIC, TIFF, WebP, GIF and BMP. Folders are indexed in place and read-only, so there is no import step and your files never move.
Get started free

Type it. Find it.

Download PhotoCore free and index up to 200 images to try search by meaning on your own archive. Full access is a one-time purchase of 59.99 euro, or 29.99 euro Early Bird until July 15.