How to Find People in Multiple Videos — Cross-Video Face Search
When you need to find the same person across many video files, searching each one individually is not an option. Here is how cross-video face search works — one reference photo, one index, results spanning every file.
Published 2026-10-08 5 min read
- video search
- cross-video
- large libraries
Finding someone in one video is a solvable problem. Finding the same person across dozens — or hundreds — of video files is a different one. You cannot open each file, scrub through it, and check. You need a way to search all of them at once, with a single reference photo, and get results that tell you which files contain the person and where.
That is what cross-video face search does. You build one index across all your video files, give it one reference photo, and the search spans everything.
The problem with searching videos one by one
Consider a typical video archive: years of footage spread across external drives, project folders, camera cards copied to laptops. Hundreds of files, in a directory tree several levels deep.
Now suppose you need to find every appearance of a specific person across all of it. The manual approach — open each file, watch for the face, note the timestamp — does not scale. Even at a minute per file, a few hundred files means hours of tedious work, and you will miss things.
What you need is a search: one query, results from every file, each result pointing to the exact moment.
How cross-video face search works
SnapByFace treats your entire video collection as a single searchable space. The process has two phases:
Phase 1: Build a unified index
Point SnapByFace at one or more folders. It walks them recursively, finding every supported video file (MP4, MOV, AVI, MKV, M4V, WebM, TS) and every supported photo (JPG, PNG, BMP, TIFF, WebP, HEIC/HEIF).
For each file, it samples frames at the configured interval, detects faces, and encodes them. All of these encodings go into a single index on your machine — regardless of which file they came from.
The index does not store images. It stores numeric face embeddings — the same kind of vector the app uses to compare faces in photos. The original files are never modified.
Phase 2: Search once, find everywhere
Add a reference photo of the person you are looking for. SnapByFace encodes that face and compares it against every face in the unified index. The results come back grouped by file, each with a time range:
interview_2024_03.mp4— 02:14–02:34, 15:07–15:22b_roll_park.mov— 00:43–00:51event_recording.mkv— 01:12:05–01:12:18
One search. Every file. Every moment.
Setting up for multi-video search
Organize your folders (or don’t)
SnapByFace walks directories recursively, so you do not need to flatten your folder structure. Point it at the root of your video archive and it will find everything inside.
That said, if you have multiple storage locations (an internal drive, an external SSD, a NAS), add each one as a separate watched folder. The app handles them all in the same index.
Add your reference photos
The people library works the same way as for photo search: one clear photo per person, and the file name becomes the person’s name. The same library is used for both photo and video results.
If you have multiple photos of the same person (different angles, different time periods), you can add them all. SnapByFace uses all of them when matching.
Start indexing
Indexing runs in the background. For large collections, it takes time — but it only needs to happen once. After the initial index, new files are picked up incrementally.
You can pause, resume, and cancel indexing at any time. Progress persists across app restarts.
Managing large video libraries
Incremental indexing
When you add new video files to a watched folder, SnapByFace detects them and indexes only the new content. It does not re-analyse files that are already in the index.
Similarly, if a file is modified (different size or modification time), it is re-indexed. Unchanged files are skipped.
Performance at scale
The search itself is fast regardless of library size. Looking up a face in an index of 100,000 faces takes the same order of magnitude as looking it up in 1,000. The bottleneck is always the initial indexing — the frame sampling and face encoding — not the search.
For very large collections (tens of thousands of video files), the practical approach is:
- Start with your most important folders first.
- Let the initial indexing finish (or run it over a weekend).
- Add more folders gradually. The app handles them all in the same index.
Pausable, resumable scans
Hours of 4K footage takes time to index. If you need your machine’s resources, pause the scan. Close the app. Resume the next morning from exactly where it stopped.
This also protects against unexpected events — a crash, a power loss. The progress is persisted, and the scan picks up where it left off.
Use cases
Journalism and documentary
You have hundreds of hours of raw footage across multiple shoots and need to find every appearance of a specific subject. Cross-video search gives you that in minutes instead of days.
Legal and compliance
You need to locate every moment a specific individual appears in a set of recordings. The search provides time-stamped results across all files, which you can then review and export.
Video production
You are editing a documentary or compilation and need to find the best takes of a specific person across multiple recording sessions. One search, results from every file.
Personal and family archives
Years of family video spread across devices and drives. You want to find every moment a specific family member appears — across all of it, in one search.
What stays on your machine
Everything. Recognition, indexing and search all run locally. No video, no frame and no face data is uploaded. The index lives in ~/.snapbyface/ and contains only numeric embeddings, not images.
SnapByFace reads your media and never writes to it — no renaming, no moving, no re-encoding.
The only outbound data is an optional device statistics ping (machine code, app version, OS, language, activation state — no filenames, no footage, no search activity). On by default, turnable off in Settings.
Supported formats
Video: MP4, MOV, AVI, MKV, M4V, WebM, TS. Photos: JPG, JPEG, PNG, BMP, TIFF, WebP, HEIC/HEIF.
Camera RAW files (CR2, NEF, ARW and similar) are not supported. Photos and video share the same index, so one reference photo finds a person across both media types.
Try it on your collection
The trial runs the full local search and shows up to 5 matches per person and 20 per search — enough to verify that it finds the right person across your files. During the trial you can start up to 100 analyses in total.