Hide faces and objects in video, on your device
macOS app that hides faces, plates and any other object in a live camera feed or a saved video or image, using an on-device YOLO26 segmentation model with object tracking. Nothing is ever uploaded.
v1.0.0 · Apple silicon & Intel · macOS 14 Sonoma or later · release notes
Install
- 1Download the .dmg and open it. Redact runs inside apps-releases, the launcher these apps ship in. Redact is the first.
- 2Drag apps-releases into Applications.
- 3Right-click apps-releases → "Open" → "Open" once, to approve it (it's ad-hoc signed, not notarized). Redact opens straight away.
If macOS blocks the app, select Open Anyway under System Settings → Privacy & Security.
Built for privacy first
Runs entirely on-device
A bundled YOLO26n segmentation model runs on the Neural Engine. No network calls, no accounts, no telemetry, no cloud service. Your video never leaves the Mac it's running on.
Live camera or saved files
Redact a webcam feed as it runs, or drop in a video or image: see it untouched, apply the redaction, then export. Processed frames appear as they encode, so you watch the work happen.
Tracking keeps the blur glued on
A per-frame detector is not stable, and a single missed frame is an unblurred face in the export. Redact follows each object between frames, so a blur survives a dropped detection and a pass behind someone else.
Pick exactly what to hide
Choose any of the 80 objects the bundled model knows (people, license plates, laptop screens, phones), or bring your own Ultralytics model to detect something else entirely.
Box or mask, pixelate or blur
Box covers the whole rectangle; Mask clips to the object's own outline so the background inside it stays sharp. Both come as sharp pixelation or a soft blur, with an intensity slider.
Frequently asked questions
- Is Redact free?
- Yes. There is no sign-up, no account and no limits.
- Is my video uploaded anywhere?
- No. Detection and blurring both run locally via CoreML, on the Neural Engine or GPU. Nothing is ever sent to a server.
- What can it hide besides faces?
- Any of the 80 object classes the bundled YOLO26 model recognizes (license plates, laptops, phones, TVs and more), selectable individually or all at once. You can also load your own Ultralytics model to detect something the bundled one does not know.
- Can it process a video file, not just the camera?
- Yes. Drop in a video or image, look at it untouched, apply the redaction and export the result. Video keeps its audio, and the processed frames appear while it encodes so you can see what you are getting.
- Does it slow down my Mac?
- It runs a nano segmentation model on the Neural Engine, built for real-time inference on a live camera feed. Box-method blurring is the fastest option if you want maximum frame rate.
- Is the download safe if it's not notarized?
- The .dmg is ad-hoc signed rather than notarized through the App Store, so Gatekeeper blocks it on first launch. Right-click the app → "Open" → "Open" once to approve it. This is a one-time step for any app distributed outside the App Store. If macOS blocks the app, select Open Anyway under System Settings → Privacy & Security.
- Does it slow down as the video gets longer?
- No. Frames are processed and written as they stream, so memory stays flat whether the clip is ten seconds or ten minutes.