Privacy First
Face recognition is an optional, consent-gated feature. Models download to your machine and run locally. Embeddings are encrypted at rest and never uploaded. When you share an index, biometric data is excluded by default.
InsightFace Buffalo-L runs via ONNX with CUDA and Metal acceleration. No biometric data is ever sent to our servers.
Face embeddings and thumbnails are encrypted at rest using Age encryption. Only accessible when the app is running.
Nothing runs until you explicitly opt in with two separate consent checkboxes. Full audit trail of consent events.
BIPA, GDPR, and CCPA compliant. Biometric data is wiped on logout and excluded from shared indexes by default.
How It Works
Opt in through the consent dialog. Two checkboxes: processing consent and data retention acknowledgment. Models download automatically.
FrameQuery extracts key frames, detects faces, and generates 512-dimension embeddings. Similar faces are clustered automatically.
Link face clusters to named people. Merge duplicates with one click. Create new people on the fly from any detected face.
Type @Sarah in the search bar to find every appearance. Autocomplete suggests matching people as you type.
People Management
Once you name a person, FrameQuery links that identity across your entire library. New videos with matching faces are tagged automatically.
Merge duplicate detections, rename people, or delete them entirely. Changes propagate across every video in your library instantly.
Type @ followed by a name in the search bar. For multi-word names, use @"full name".
Results show every video and scene where that person appears, ranked by relevance. Combine with other search terms to narrow results.
Voice Recognition
Voice recognition works alongside face detection. Speaker embeddings (ECAPA-TDNN) identify unique voices from the audio track. If a face and voice appear in the same frame, FrameQuery links them to the same person automatically. Filter transcript results by speaker.
Available on every plan including Free. Enable when you're ready and FrameQuery backfills any previously indexed videos.