How to find photos by describing them
Updated 20 October 2026
To find a photo by describing it, you need image search that looks at what a picture shows, not at its file name. Windows 11 offers that through Microsoft’s semantic search, including its Photos app, but only on Copilot+ PCs. The Mac Photos app can do it with Apple Intelligence, and for photos sitting in ordinary folders on any supported PC or Mac, BeaconFind searches them by content on your own computer.
Somewhere in eight thousand files named IMG_2039.jpg and Screenshot 2024-11-02.png is the one picture you need right now, the whiteboard from that meeting, the receipt you photographed, the sunset from the lake trip. File names won’t help; they never contained any information to begin with. What works is searching by what’s in the image.
What the built-in tools can and cannot do
Windows. File Explorer finds pictures by file name, date and other properties. For photos, sort by Date taken rather than Date modified, since copying a file changes the modified date but not the date the photo was taken. That narrows a folder quickly when you remember roughly when, and it says nothing about what the picture shows. Microsoft does have content-based photo search: its Photos app can find locally saved, indexed photos in your Pictures Library from a description such as “sunset at the beach”. Microsoft documents that for Copilot+ PCs, which are defined by a dedicated NPU, so a PC without one does not have it. The details are in Windows 11 semantic search without a Copilot+ PC.
Mac. Apple’s Photos app searches titles, captions, keywords, dates and even text within photos, and with Apple Intelligence you can type a description in natural language; Apple notes that this is not available in every language or region. That is a search inside the Photos app. For the file-level side, such as pictures kept loose in folders, see how to search inside files on Mac.
What “describing a photo” actually means
Modern vision models can place images and text in the same “meaning space”: the photo of a dog on a sofa ends up close to the words “dog on a sofa”. A search tool that uses this lets you type a description and ranks your local images by how well they match. No tags, no albums, no manual sorting. BeaconFind’s photo search uses CLIP, a model built for exactly this pairing of pictures and words.
Describe the scene the way you would to a friend:
- Scene and setting: “mountain sunset”, “snowy street at night”.
- Objects and colours: “red bike at the beach”, “birthday cake with candles”.
- Things that look distinctive: “whiteboard with diagrams”, “handwritten notes”.
Two tips make a difference. Say photo or picture in the query, which tells the search you want images rather than documents, and open the Images tab to see the results as a grid of thumbnails instead of a list. And describe what you can see, not what you remember about the occasion: a search for “the planning meeting” says nothing about how the whiteboard looks.
OCR versus image understanding: two different jobs
These get confused, and they work differently. Image understanding is what finds a picture by what it shows. OCR (optical character recognition) reads the letters inside an image and turns them into text, which is how a screenshot of an error message or a booking confirmation becomes searchable. Many PCs hold thousands of screenshots, where the text is pixels, not text, which is why normal search cannot see it.
There is a limit worth knowing. BeaconFind runs OCR over screenshots saved as PNG, WebP, BMP or TIFF and over scanned PDFs. It does not run OCR over JPEG photos, so the words on a photographed receipt saved as a JPEG are not read. For those, search by what the picture shows, or by the date. Scanned pages are covered in how to search scanned documents.
On the Balanced and Max tiers BeaconFind also writes a short description of each photo, which becomes searchable text beside the visual match. The lightest tier keeps the photo search but not the written descriptions.
Where it falls short
- Look-alikes. The search ranks by closeness, so a similar picture can appear beside the exact one.
- Counting and fine detail. Models like this are good at scenes and objects and weaker at exact counts, small print or telling two similar people apart.
- First index. Every image has to be processed once, and the AI models download once on first run, so a large photo library takes time before search is complete.
Why on-device matters most for photos
Photos are the most personal files you own. Cloud photo services can search by content too, but only by uploading your pictures to their servers first. The alternative is to run the image-understanding models locally, so the search works the same but your photos stay on your computer. Your files, their contents, your searches and the index never leave your machine; the only network traffic is a licence check and an update check, which you can block at your firewall without affecting search.
Doing it with BeaconFind
BeaconFind brings these together on your own PC or Mac: visual search (describe the scene), written descriptions on the higher tiers, and OCR for screenshots and scans. Press Ctrl+K on Windows or Option+Space on a Mac, type “whiteboard with diagrams” or “red bike at the beach”, and jump straight to the picture, even in a folder of ten thousand unnamed images. If you cannot remember anything about the picture at all, start from how to find a file when you forgot its name, and for the idea behind it see what semantic file search is.
Find that photo in seconds, free for 14 days, no card. Runs entirely on your PC.
Download BeaconFindFrequently asked questions
How do I find a photo on my computer by describing what is in it?
Use a search tool with image understanding. It converts your photos and your description into the same kind of numbers, so a description like red bike at the beach ranks the matching pictures first even though the file is called IMG_2039.jpg. Windows 11 does this in the Photos app on Copilot+ PCs, the Mac Photos app can with Apple Intelligence, and BeaconFind does it for photos in ordinary folders on any supported PC or Mac.
Can Windows search the content of photos?
File Explorer searches names, dates and other properties, not what a picture shows. Microsoft's semantic search in the Photos app can find locally saved, indexed photos in the Pictures Library from a description, but Microsoft documents it for Copilot+ PCs, which need an NPU above 40 TOPS.
Can it find text inside a photo?
BeaconFind reads text out of screenshots saved as PNG, WebP, BMP or TIFF, and out of scanned PDFs, using OCR. A JPEG photo of a receipt or a whiteboard is not run through OCR, so find those by describing what the picture shows instead of the words on it.
Does photo search work on a Mac?
Yes, in two ways. Apple's Photos app searches titles, captions, keywords, dates and text in photos, and with Apple Intelligence it accepts natural-language descriptions, subject to language and region availability. BeaconFind runs on Apple Silicon Macs and searches photos stored in any folder, not only the Photos library.
Are my photos uploaded anywhere?
Not by BeaconFind. The image-understanding models run on your own computer, and your files, their contents, your searches and the index stay there. The app contacts our servers only for a licence check and an update check, and both can be blocked at your firewall without stopping search.