Blurring faces in photos: what automatic detection can and cannot do

An automatic face detector can save you drawing dozens of boxes. It cannot tell you that everyone in a photograph is hidden. The useful workflow is to let it mark likely faces, correct the regions, then inspect the file you will actually share.

What the boxes tell you

Open the photo in Face Blur and wait for detection to finish. The tool returns face regions and confidence scores. It does not identify the people in the photograph. A high score means the detector considers that region likely to contain a face; it does not rate the strength of the blur or the completeness of the coverage.

Returned detections below 50% are marked needs review, but unflagged boxes need checking too. If the result says No faces detected, you can still draw regions manually.

Review the photograph in two passes

First look at the people the tool has marked. Check the forehead, ears, chin and any face details extending beyond the treated shape. Then scan the rest of the photograph without using the boxes as your guide:

• Look between people in a group and behind the main subject.
• Inspect small faces near the edges of the frame.
• Check profiles, faces looking down and faces partly hidden by hair or hands.
• Look in mirrors, windows and screens for reflected or pictured faces.
• Review poorly lit and motion-blurred areas rather than assuming they contain nothing recognisable.

These are places to look, not a list of faces the detector will necessarily miss. In Single image mode, drag on an empty area to add a region. Drag an existing region to move it, or use Bigger and Smaller in its list entry. A region set to Skipped receives no effect.

Padding and shape solve different problems

Region padding enlarges the detector’s box. At the default 35%, a box 100 pixels wide gets 35 pixels added on each side, becoming 170 pixels wide away from the image boundary. This is a proportion of the face box, not a fixed margin for every photo.

Region shape determines which part of that box is treated. An Ellipse does not cover its corners. Rounded rectangle also leaves curved corners; Rectangle covers the full box. A generous outline can therefore still leave a piece of a face showing if the shape is a poor fit.

Set padding before making manual corrections. In the current editor, changing it can restart detection and replace those corrections. Detect again also replaces the region list. After either action, check that the regions you need are still present.

Choose between softened detail and opaque cover

Blur softens detail; Pixelate creates a mosaic. Both use information from the original image. Neither an obvious effect nor a high strength setting proves that a person is unrecognisable.

Solid cover paints an opaque colour inside the selected shape. Use it when the facial detail needs to be removed from the output, and check the coverage around the edges. It changes only that region: a name badge, tattoo, clothing, location or accompanying caption may still identify someone. Other visible details can be covered with image redaction.

A batch still needs individual review

Batch mode accepts up to 100 JPEG, PNG or WebP files, each up to the tool’s 25 MB limit (25 MiB). Both modes reject images above 40 megapixels. Batch processes files one at a time and offers a ZIP of completed results, including images where no faces were found. Completion tells you processing finished, not that the photograph passed a review.

The batch Review button opens the original in Single image mode. If you add or change regions there, download the corrected image separately. Those edits do not replace the version in the batch ZIP. When gathering the files to share, make sure you use the corrected download.

Check the export, filename and context

Choose Download blurred image after reviewing the regions; the button keeps that name even when you use Solid cover. Open the downloaded image at full size. Check the placement and strength of each effect, then make one more sweep for missed faces. Do not assume a small preview is an exact representation of the exported treatment.

The tool processes images in the browser, using a detector and runtime served from ToolsNow. Export draws the result onto a canvas and encodes a new file, so the original EXIF metadata, including any GPS tags, is not copied across. You can inspect the download with the photo metadata viewer; the new encoder may add its own file fields. The original photo on your device is not overwritten.

The download keeps the original filename’s stem and adds -faces-blurred. A person’s name or a location in that filename is still worth removing. Check captions and the scene itself too: removing face detail and metadata does not establish anonymity.

Last reviewed 2 October 2026.