Hide every face
entirely on your device

Apple Vision finds the areas to cover. Core Image draws the covers, and AVFoundation reads and writes video. Garim uses Apple’s frameworks, with no company server or cloud AI.

Face detection and covering work with the network off. Photos and videos stay on the device.

Apple Vision locates face rectangles. High sensitivity adds a search in smaller tiles; Max adds a centre tile and runs the tile search on every frame. Overlapping detections are merged.

For front and profile faces that face detection misses, body pose estimates the head area from the nose, eyes and ears, and Garim covers it too.

For plates, name tags and signs, text detection uses Korean and English. Garim uses only the positions of the text boxes and covers each as a full rectangle. Precision mode adds more detection work, padding and smoothing.

Vision returns positions. Garim follows people by position and motion, without matching their appearance or identity. No face data is stored.

In the analysis pass, every decoded frame gets a full-frame face search. Tile and pose searches run at the selected sensitivity; text detection runs on every frame when enabled.

Predicted position and box overlap link detections into tracks. Interpolation fills short gaps of 0.7–1.1 seconds, or 1.2 seconds in precision mode. Tracks extend 0.3–0.6 seconds at each end, and their boxes are smoothed over time.

In the rendering pass, Garim applies the tracked covers to every frame and writes a new video. The original audio is added back without re-encoding.

The face box expands by 12% on each side for video and 10% for photos. Circles and ovals enclose the box’s corners; their edges are anti-aliased rather than feathered.

Mosaic block size follows the largest covered area. Its width ÷ (10 − 6 × strength) gives about 10, 7 or 4 blocks across the largest covered area at 0%, 50% or 100% strength. Every block uses the exact average colour of its pixels.

Blur uses a Gaussian radius of face width × (0.10 + 0.15 × strength), or 17.5% of face width at 50% strength. Solid colour fills the area with the chosen colour; stickers keep the proportions of your image or emoji.

The cover is edge-clamped before averaging or blurring. HDR video is processed in its HLG/PQ values; an HDR photo receives a new brightness map calculated from the covered result.

Block sizes in the original 1920 × 1080 image.

The saved photo or video is flat, with no removable mask layer. Video keeps its pixel size and frame rate, using H.264 by default, optional HEVC, or 10-bit HEVC for HDR.

Photos are saved as PNG or, for HDR sources, HEIC. Saved photos carry no EXIF or GPS metadata from the source.

Garim reads a temporary copy of the selected video and saves the result as a new file. Temporary files are cleared when a job finishes or is replaced.

History keeps the result, its PDF record and a thumbnail until you delete it. It does not keep the original.

Select photos and videos through the system file importer, including several at once. App Sandbox limits access to the files you select. A save panel writes the result as a new file.

Mac uses the same processing engine and the same purchase as iPhone and iPad.