Face blur hides faces before you share a photo: as soon as an image is added, face-landmark detection runs and every face box is expanded by 30% (to cover the hairline and chin). You can then toggle each box, drag it, pull its corners to resize it, or add missed faces by dragging on empty space or clicking Manual box. Three methods are available — mosaic (block size as a percentage of image width), blur (a padded neighborhood is blurred as a whole and pasted back so edges do not darken) and a solid-color cover — and the preview and the export share the same settings. Everything runs locally in the browser.

Unlike Pixelate in the workbench, which is manual brushing of arbitrary areas, this tool finds faces automatically, uses rectangular boxes and works on several images at once.

1. 01Add images— drop in several; each thumbnail badge shows how many boxes were detected, and clicking a thumbnail switches to that image.
2. 02Check the boxes— under the preview, “N faces detected, x/y boxes enabled” is shown; click a box to select it, then drag it or pull its corners, and use the small buttons on the box to disable or delete it. Dragging on empty space creates a new box (accidental drags smaller than 12 px are discarded).
3. 03Pick a method— Mosaic with a block size (0.5–6% of image width), Blur with a strength (0.5–6% of image width) or Cover with a color.
4. 04Blur and export the batch— the button shows the number of images and is marked irreversible, and it is blocked when no box is enabled. Files are named name_blurred.ext with a note of “N blur areas” and can be packed as face-blur.zip.

Input

— JPG, PNG, WebP, GIF, BMP, AVIF and HEIC; the output keeps the original format (GIF and BMP become PNG) at JPEG/WebP quality 0.92.

Detection limit

— up to 6 faces per image, with the detection input downsampled to 1,600 px; very small or heavily turned faces can be missed and need a manual box.

Working resolution

— images longer than 4,096 px on the long edge are downsampled first, so the export resolution drops with them.

Model load failure

— if the wasm or model files cannot load, the tool switches to manual mode and you can still drag boxes.

Irreversible

— mosaic, blur and cover rewrite the pixels and cannot be undone, so keep your own copy of the original.

What it does not do

— it does not blur number plates, text or other non-face targets (draw a manual box instead), does not do freehand brushing (use Pixelate) and does not process video.

Photos are processed in your browser throughout and are never uploaded to any server; the model and wasm files needed for face detection are loaded from this site's own path on first use (about 15 MB, and the request does not contain your photo), falling back to manual boxes if the load fails.