How to Find Every Photo of One Person in a Large Library

Finding every photo of one person comes down to three things: one good reference photo, a single pass over your folders, and knowing how to read a similarity score. Here is the workflow that holds up at 50,000 photos.

“Which photos is my daughter in?” sounds like a simple question until your library is large enough that scrolling is not an option. Albums do not help (they sort by event, not by person), filenames do not help, and cloud photo apps only help if you are willing to hand over every face you own.

The reliable answer is reference-based search: give the computer one good photo of the person, let it index your folders once, and ask again whenever you like. Everything below runs locally, on your own machine.

Why browsing stops working

Three things break down as a library grows:

Time-based browsing works when you remember roughly when a photo was taken. Beyond a few thousand photos, you stop remembering.
Album or folder organisation sorts by event. A person appears across dozens of events, so the answer is scattered by construction.
Manual tagging is accurate and completely unsustainable — nobody tags 30,000 photos.

Person-based search flips the direction: instead of remembering where a photo is, you describe who you are looking for and let the index answer.

Start with one good reference photo

The quality of your reference photo sets the ceiling for everything after it. One clear photo beats five mediocre ones.

A good reference photo is:

Front-facing, or close to it. A profile shot gives the model roughly half the information.
Well lit and in focus. The face should occupy a decent part of the frame.
Recent enough to resemble the person across your library. Faces change; a photo from 20 years ago is a weak reference for photos from last month.
One face. A group photo forces you to crop, and a cropped face is usually a worse reference than a dedicated portrait.

If a person looks very different across your library — a child growing up, or someone with and without glasses — use two or three reference photos of that person. SnapByFace treats all reference photos of a person as the same identity.

The workflow

Add the person. Point SnapByFace at a folder of reference photos, with each file named after the person — the file name becomes their name.
Add your media folders. Point it at the folders holding your photos. SnapByFace walks them recursively, so you do not have to flatten anything first.
Let it go through the folders once. It looks at every photo and remembers every face it finds. Progress is visible and the work can be paused and resumed.
Review the matches. Results come back sorted by similarity, per person or per file. Double-click a hit to open the preview.

After the first pass, re-indexing is incremental: adding new photos does not re-analyse the old ones.

How to read a similarity score

Every hit carries a similarity score: how close the face found is to your reference photo. SnapByFace reports a hit once that score reaches the threshold — 0.45 by default, adjustable anywhere from 0.30 to 0.80.

Two practical rules:

Raise the threshold (toward 0.60+) when you are getting too many wrong people. Stricter means fewer, more confident hits.
Lower it (toward 0.35) when you are missing people you know are there — useful for profiles, motion blur, and faces partly turned away.

:::tip Changing the threshold never re-analyses your media. Your folders have already been looked at once; a new threshold is just a stricter or looser cut-off over the same findings. :::

Working with the results

Results are viewable by person (everyone found for one reference) or by file (everyone found in one photo). Both views are useful for different questions: “where does she appear?” versus “who is in this photo?”

When you need the list outside the app, export the results to CSV. And if you want to check a hit against the source, double-clicking opens the preview — SnapByFace never renames, moves, or modifies your files, so your library is read-only to it.

Where the work happens

All of it runs on your own computer. Your photos and videos stay there: everything the app learns about them is kept in ~/.snapbyface/, and no image or video is uploaded. There is no account to sign into.

The only thing that leaves the machine is an optional device statistics ping — machine code, app version, operating system, language and activation state. It contains no photos, no face data, no file paths and no search activity, and you can switch it off in Settings.

When this approach is the wrong tool

Be honest about the limits, because they are real:

No reference photo, no results. SnapByFace only finds people you have added. It does not do open-ended “who is this?” identification.
Very low resolution or heavy occlusion degrades any face search, local or cloud.
Huge appearance changes across decades will need several reference photos per person.

If the answer you need is “find this person and I have one decent photo of them”, local reference search is exactly the right tool. If you need “identify this stranger”, no consumer tool — cloud or otherwise — should be doing that for you either.

Try it on your own library

The free trial runs a complete local search against your own folders and shows up to 5 matches per person and 20 per search — usually enough to tell whether the default threshold fits your material. Activating unlocks every match already found; you do not re-run anything. During the trial you can also start up to 100 analyses in total.