Pixelate faces in a photo

Why pixelate rather than blur

A pixelated face is obviously, deliberately hidden. Viewers read it at once as privacy protection rather than a soft-focus effect, which is why it is common in documentary and news images. It also survives being shrunk: a mosaic still looks like a mosaic in a small thumbnail, while a light blur can start to look like an ordinary out-of-focus face.

If you prefer a softer look, Blur faces uses the same detector with blur as the main style. Both pages also offer a solid block.

Pixelation also suits screenshots of public places and thumbnails, where the mosaic stays readable as deliberate at any size.

Choosing a block size

Block size decides how much of the face survives. With small blocks, the rough shape of the eyes, nose and hairline can still come through, and someone who knows the person may recognize them. Larger blocks leave only a few patches of skin tone and hair color.

A useful rule: if a face box is around 100 px across, aim for blocks big enough that the face is only a handful of blocks wide. Faces of different sizes in the same photo need checking, because a block size that hides a close face completely may leave a distant one barely changed. When in doubt, go larger, or use a solid block.

Making sure every face is covered

The face finder, a 230 KB model, boxes every face-shaped region it finds. It never identifies anyone; it has no concept of who a person is. A Sensitivity slider controls how readily it marks a region. Raise it for crowds and distant faces. Lower it if it marks patterns or objects.

Click a box to turn it off, and drag on the photo to add a box where a face was missed. Look closely at faces in profile, faces partly hidden by hats or hands, and small faces near the edges. Those are the ones a detector is most likely to skip.

Reflections in windows and mirrors can hold faces too.

The exported file

The mosaic is applied to the exported pixels and cannot be undone from the file. The original detail under each box is not in it. Exports are re-encoded, so location and camera details from the original photo are left out too.

Keep the untouched original somewhere private, since the shared version cannot be turned back. Check the export at full size once more before posting; a missed face is hard to take back once a photo is online.

What it will not do

• Small, turned or covered faces can be missed. Check every face before you share.
• It never identifies anyone; it only finds face-shaped regions.
• Small blocks can leave a face recognizable to people who know the person.
• Only faces are boxed automatically; badges, plates and other details need boxes drawn by hand.

Devices and downloads

The face finder is a 230 KB download, MIT licensed, that starts after one press of the main button and stays cached. The editor shows its size first. It runs in a background worker, on WebGPU when available and on WebAssembly otherwise; at this size it finishes in a fraction of a second even on WebAssembly. The mosaic is ordinary image arithmetic.

On-device AI: A small neural network that runs on this device after a one-time download. See what is stored on this device.

Step by step

1. Open the photo. Drop, paste or choose the image with faces to hide.
2. Check the boxes. Set Sensitivity, turn off boxes you don't want, and drag to add missing ones.
3. Set the block size. Choose pixelate and increase the block size until each face is unreadable.
4. Review and export. Check every face at full size, then save.

Questions about pixelate faces