Face recognition for searchable video libraries
ClipCatalog detects faces in your videos, groups recurring people, and turns them into reusable person filters across your library. This page covers the broader capability behind person-based discovery across projects, archive drives, and long-running footage collections.
Need the task-first workflow for locating one person quickly? Start with how to find a person in video.
Group recurring people across a library
ClipCatalog turns repeated face detections into reusable person groups, so the same guest, subject, collaborator, or family member becomes a search dimension instead of a memory problem.
Pivot from one clip to many related clips
Start from one useful result, then expand to every matching clip that contains the same person across projects, drives, and years of footage.
Keep face processing local and controllable
Face detection, embeddings, and grouping run on your computer. Nothing is uploaded, and the resulting face data stays in your encrypted local database and FAISS index.
How person-based discovery works in ClipCatalog
ClipCatalog uses on-device AI models to detect faces and compute embeddings, then compares them in a local FAISS index to group recurring people. The result is a reusable person filter that helps you revisit the same subject across an archive.
Enable face detection
Toggle face detection on in Settings. A confirmation dialog explains the feature and its implications before enabling it.
Faces are detected & grouped
During processing, ClipCatalog detects faces using YuNet and computes embeddings with SFace — both running locally via OpenCV DNN. Similar faces are clustered into person groups.
Filter by person across the archive
Select one or more people in the search panel to see matching clips across your indexed library. If you select multiple people, switch All or Any matching and combine with tags, transcripts, dates, and other filters.
See face search in action
Start from a clip, ask ClipCatalog to find more videos with the same people, then tighten the results when you need clips where two selected people appear together.
* Some faces are pixelated for privacy.
Best for
Best when the same people appear again and again across large libraries, recurring shoots, or multi-year archives where person filters become reusable.
Archive-scale workflows
Reuse person groups in family archives
In large family archives, the same people show up across holidays, birthdays, and trips for years. Grouping recurring faces turns those appearances into a reusable retrieval workflow instead of a one-off search.
Track recurring guests across series
If the same collaborator, guest, or on-camera expert appears across many episodes, person filters let you revisit every appearance without remembering which project folder or shoot day it came from.
Search long-running interview archives by subject
Running a long-form interview series or documentary archive? Filter by a guest's face to pull appearances across seasons, then combine it with transcript search to find the exact moments they said what you remember.
Review event footage by person
Large events produce hundreds of clips with the same core people appearing again and again. Person filters help you narrow from the whole event archive to the people that matter for the edit.
What to expect from face recognition
Grouping, not perfect ID
Face recognition groups similar-looking faces so you can find clips by person. It uses k-NN voting with thresholds and margins to reduce false matches, and caps embeddings per person to avoid bias from overrepresented faces. Occasional mis-groups can happen with blurry, partial, or heavily obscured faces.
Works from thumbnails
Face detection runs on sampled frames (thumbnails), not every single frame of video. This keeps processing practical for large libraries while still catching the most visible face appearances. Very brief on-screen moments may not be detected.
Optional GPU acceleration
Face detection and embedding use OpenCV DNN, with optional OpenCL acceleration when available. If your system doesn't support OpenCL, the app falls back to CPU automatically — slower, but it still works. Learn about GPU acceleration →
Combine with other filters
Face search is most powerful when layered with other ClipCatalog filters. Find clips where a specific person appears and a keyword was spoken, and a visual tag matches, and it's from a particular date range or folder. Explore all search filters →
If face detection is disabled
When Face Detection is off, the “Footage type” filter won’t be available, and Highlight Score uses fewer signals — so it may be less accurate. You can re-enable Face Detection in Settings at any time. Learn about search filters →
When and how you may use face recognition
Face recognition in ClipCatalog is designed to be transparent and fully under your control. Before enabling it, the app shows a confirmation dialog with the following statement:
Settings view showing face detection before and after enabling it. Face search stays opt-in, and you can turn it off or delete face data later.
"Analyze faces in my videos locally to group similar faces so I can find videos with the same person.
This may be considered biometric data and can be regulated in some countries.
I am responsible for using this feature in compliance with applicable laws.
I can turn it off at any time and clear existing face data."
Opt-in only — disabled by default
Face detection is off when you first install ClipCatalog. It only activates when you deliberately enable it in Settings and confirm you understand the implications. You can turn it off again at any time — existing face data is preserved until you choose to delete it.
Biometric data considerations
Face embeddings — mathematical representations of facial features — may be classified as biometric data under laws like the EU's GDPR, Illinois' Biometric Information Privacy Act (BIPA), or similar regulations in other jurisdictions. Because ClipCatalog processes everything locally on your machine and never uploads face data, you retain full control. However, you are responsible for ensuring your use complies with the laws that apply to you.
Local processing — no cloud involvement
All face detection, embedding computation, and identification happen on your computer. Face crops, embeddings, and the FAISS index are stored locally in your encrypted ClipCatalog database folder. No face data is ever sent to a server or cloud service.
Full data control — delete anytime
You can delete all face data at any time from Settings: face images, detections, person groups, and the FAISS face index are all removed. If you re-enable face detection later, videos will need to be reprocessed. This gives you a clean way to remove all biometric data if you change your mind or if your circumstances change.
Even more powerful together
Face recognition is one search dimension in ClipCatalog. Its real value is combining person filters with the rest of your archive metadata to go from thousands of clips to exactly what you need.
Find clips by what was said — perfect for interviews, sound bites, and voiceover takes.
Search by what's on screen — scenes, objects, and actions, tagged automatically.
Face data persists even when drives are unplugged — reconnect and search again.
Layer face filters with date, folder, resolution, frame rate, duration, and more.
Relevant comparisons
If you are evaluating this workflow against other tools, start with these side-by-side pages.
Frequently asked questions
No — face recognition is completely optional. You can use ClipCatalog for detected content, transcript search, and all other features without ever enabling face detection.
No. Face detection, embedding, and grouping all happen locally on your computer. Your footage and face data never leave your machine.
Open Settings and toggle "Face Detection" on. You’ll see a confirmation dialog explaining what the feature does, that face data may be considered biometric data in some jurisdictions, and that you can turn it off and delete face data at any time.
Yes. In Settings you can delete all detected faces, person groups, and the face index with one click. Videos will need to be re-processed if you enable face detection again later.
Yes. You can layer face/person filters with transcript words, detected content, date ranges, folders, resolution, frame rate, and more — so you can find exactly the clip you need.
ClipCatalog uses a smart matching algorithm designed to reduce false matches and avoid bias from overrepresented faces. It’s built to get you to the right set of clips quickly, though occasional mis-groups can happen — especially with low-quality or partial faces.
Face detection and embedding use OpenCV DNN with optional OpenCL acceleration. If OpenCL isn’t available on your system, it falls back to CPU automatically.
In many jurisdictions, face embeddings and facial geometry are classified as biometric data and may be subject to privacy regulations such as GDPR (EU), BIPA (Illinois, US), or similar laws. ClipCatalog processes everything locally and gives you full control, but you are responsible for using this feature in compliance with applicable laws in your region.
Try it with one folder
The best way to evaluate the broader face-recognition workflow is to enable it in Settings, process a single project or archive folder, and see how quickly one useful clip turns into a reusable person filter.
Try ClipCatalog free — up to 500 videos
No account required. Your footage stays on your computer.
Understanding face recognition for video libraries
Whether you call it face search, person discovery, or facial recognition for video, the core idea is the same: let software turn recurring faces into a reusable retrieval layer across your archive.
Why person-based discovery matters
Sometimes the problem is not just finding one shot. It is revisiting a recurring guest, a documentary subject, or a family member across a large archive. Face recognition turns "I know they appear throughout this library" into a direct filter by person, which is much broader than a one-off manual search.
Local vs. cloud face recognition
Most face recognition services require uploading footage to a cloud API. For video creators working with client footage, personal content, or simply large files, that's often a non-starter. ClipCatalog runs all face processing on your own hardware — your footage and face data stay on your machine. More about local-first privacy →
How face embeddings work
When ClipCatalog detects a face, it computes an embedding — a compact mathematical representation of that face's features. These embeddings are stored in a local FAISS index, making it fast to compare new faces against all previously seen ones. Similar embeddings are grouped together as the same person, which is how search-by-person works.
Practical limitations
Face recognition works best when faces are clearly visible, well-lit, and facing roughly toward the camera. It can struggle with distant faces, heavy shadows, motion blur, or extreme angles. The system works from sampled frames, so very brief appearances may be missed. Knowing these limitations helps you search smarter and set realistic expectations.