Blur a face in a photo
Blur faces before you post a photo. Drag a box over each face, choose blur, mosaic or a solid block, and export with EXIF and GPS data stripped.
- Runs in your browser
- Files are never uploaded
- Free, no account
How to blur a face
- Drop the photo in. It is decoded on your device; no copy is sent anywhere.
- Drag a box over each face. The shape selector is set to Oval, which follows a head better than a rectangle.
- Raise the strength until the face is unrecognisable in the preview — the preview is a scaled copy of the exact pixels you will get.
- Download. The file is re-encoded from raw pixels, which drops the EXIF block, the GPS coordinates and the embedded camera thumbnail.
Why we ask you to draw the boxes
Automatic face detection needs a machine-learning model. There are only two honest ways to get one onto this page: download several megabytes of weights, or send your photo to a server that already has them. The second breaks the promise this whole site is built on, and the first is a poor trade for an operation that takes you three seconds. Manual boxes also never miss the face in profile, the child at the edge of the frame or the reflection in the window — the cases detectors are worst at, and the ones that matter most when you are publishing a photo of other people.
Blur, mosaic or block — which is actually safe?
| Effect | Best for | Recoverable? |
|---|---|---|
| Gaussian blur | Faces, tattoos, house numbers | Not in practice at strengths above ~50%: the averaged pixels no longer contain the original values. |
| Mosaic / pixelate | Faces where the effect should look deliberate | Weak for short structured text; fine for faces if the region is only a handful of blocks across. |
| Solid block | Any text — names, addresses, plates, card numbers | No. There is no information left to attack. |
The published attacks on blurred and pixelated content all work the same way: take every plausible original — every four-digit number, every name from a list — apply the same filter, and compare. That is devastating against short, predictable text and irrelevant against a human face, which has no small candidate set. Use blur or mosaic on faces; use a solid block on anything you could type.
The metadata problem nobody mentions
Blurring the face is only half the job. A phone photo typically carries GPS coordinates accurate to a few metres, the exact capture time, the device model and serial, and — crucially — an embedded JPEG thumbnail. That thumbnail is a separate small copy of the image, and in many editing workflows it is not regenerated, so the unblurred face survives inside a file that looks redacted. Because this tool re-encodes from raw canvas pixels rather than patching the original file, every one of those is discarded. If you want to inspect what a photo is carrying before you publish it, open it in the image metadata viewer.
Practical settings
- Strength is relative. It is measured against each box's own size, so a small box automatically gets a proportionally strong blur. That is why the preview and the full-resolution export match exactly.
- Cover more than the face. Hair, ears and jawline carry recognisable shape. Extend the oval a little past the hairline.
- Watch for second appearances. Mirrors, glass doors, monitors and photographs on the wall behind your subject all repeat the face.
- Check the preview at full size. Zoom the page to 100% before deciding a face is unrecognisable — a face that is illegible in a 400 px preview may not be in a 4000 px file.
When you are publishing photos of other people
Blurring faces is the standard first step for protest photography, school and club newsletters, property listings with people in shot, medical or clinical imagery, and any dataset you plan to share. Two practical additions: strip the location as described above, and consider whether clothing, tattoos, vehicles or a house number in the background identify someone just as effectively as their face. The tool lets you add as many boxes as you need, so cover all of them in one pass.
Related tools
Pixelate Image starts in mosaic mode, Censor Image in solid-block mode for documents and screenshots, and Blur License Plate is the same engine with car listings in mind. To clear metadata from a photo you do not otherwise want to change, use Remove EXIF Data.
Questions
Does this detect faces automatically?
No, and that is deliberate: automatic detection needs a machine-learning model, and the honest ways to get one are a several-megabyte download or a server round trip. You draw the boxes yourself, which takes a couple of seconds and never misses a face the model would have missed.
Can the blur be reversed?
Not from the exported file. The blurred pixels are averaged together and written into the output — the original values are gone, unlike an overlay in a PDF or a slide deck, which can simply be deleted. At strengths above about 50% even a restoration model has nothing to work with.
Is the photo uploaded to check for faces?
Never. The photo is decoded, masked and re-encoded by your own browser. Watch the network tab, or switch off Wi-Fi first — the tool behaves identically.
Does it remove the location the photo was taken?
Yes. The export is re-encoded from raw pixels, so EXIF, GPS coordinates, the camera serial number and the embedded thumbnail are all dropped. That embedded thumbnail matters: it often survives edits elsewhere and still shows the unblurred face.
What is the best shape for a face?
An oval. Set the shape selector to “Oval” before you drag, and the mask follows the head instead of leaving a hard rectangle that draws the eye.