How to blur faces in photos so they actually stay hidden

To hide faces in a photo, find every face, cover each one with heavy pixelation or a solid box rather than a gentle blur, check the whole picture by eye for faces that were missed, and share only the new copy. I built the face blur tool to do the finding automatically on your own device, because the photos people most want to anonymise, of children, patients or bystanders, are exactly the ones they least want to upload anywhere.

Blurring faces step by step

• Open the photo. Drop it on the face blur tool. It is decoded in your browser and never leaves your device.
• Press Find faces. On first use the tool downloads a small face detector (UltraFace, about 1 MB, served from this site) and the runtime that executes it (about 10 MB, from the jsDelivr CDN). It then marks each face it finds with a box drawn slightly larger than the face, to take in the hairline and chin.
• Choose the effect and strength. Blur, pixelate or a black box, with a strength slider. Each effect is applied to an oval inside the box.
• Fix the boxes. Drag across any face it missed to add a box, and click a box to remove a false detection, such as a face on a poster you are happy to keep.
• Download. The result is saved as a new PNG if you started with a PNG, otherwise as a JPEG. Because it is redrawn from a canvas, the new file does not carry over the original camera metadata such as GPS location.

Blur, pixelate or black box

The three effects look different, and they are not equally safe. A Gaussian blur replaces each pixel with a weighted average of its neighbours. That is a smooth, predictable operation, so a light blur can sometimes be partly undone with deblurring techniques, and researchers have re-identified blurred and pixelated faces by comparing them with candidate photos processed the same way. Pixelation averages the face into blocks: with small blocks a lot of the structure survives; with large blocks, where a face becomes a handful of squares, very little does. A solid box replaces the pixels entirely and is the only option that leaves nothing to work with.

So the strength matters more than the style. For anything where identification would cause real harm, I use the black box or pixelation at high strength, and I keep light blur for cosmetic jobs, such as softening a stranger in the background of a holiday picture. How face detection works, and what really hides a face explains the detection and anonymisation side in more depth.

Faces the detector misses

No detector finds every face, and it is better to know where this one struggles. It analyses a 320×240 pixel copy of your photo, so in a large group shot or a crowd, distant faces shrink to a few pixels and can be skipped. Faces in profile or looking down, faces partly hidden by masks, hands, hats or sunglasses, and faces that are backlit or motion-blurred all score lower. Detectors can also perform unevenly across skin tones and ages. Treat the automatic boxes as a first pass: zoom in, scan every part of the picture, including reflections and the edges of the frame, and drag a box over anything it left. The tool reports how many faces it found, which makes a quick count against the people you can see easy.

What else can identify someone

A hidden face does not always mean a hidden person. Look for name badges, school uniforms and crests, distinctive tattoos or hair, car number plates, house numbers and street signs, screens showing names, and faces reflected in windows, mirrors or sunglasses. Because the blur covers an oval, ears and hair at the edges of the box can stay visible; draw a larger box if they matter. For names, account numbers and other non-face details in a screenshot, the photo blur tool lets you cover any rectangle. If you edit with other software, strip location data with the EXIF remover; What your photos reveal about you explains what a photo file can reveal.

Before you share

Do the anonymising on the original, before anything is posted. A lightly blurred version that is already online is the one that gets copied and analysed, and a later, stronger edit cannot recall it. Check the final image at full size, not as a thumbnail, and share only the edited copy. If the photo shows other people's children, schools and clubs often have rules about posting, and asking a parent first is simple courtesy. Finally, remember that the whole-image blur tool is for softening backgrounds and making placeholders, not for redaction: it treats every part of the picture the same way, which is not what you want when one face needs to vanish completely.