Adjust the boxes

Free forever. No sign-up, no watermark, no limits — ads keep it that way.

The detector has put a box over every face it found. This step is where you fix what it got wrong and cover what it was never looking for. Every edit here can be undone, so nothing is permanent until you save.

Working on the image

Drag on an empty part of the photo to draw a new box. Drag inside an existing box to move it. Drag a corner to resize it. A tap or a click selects a box, and the selected box is the one the toolbar acts on. To remove it, press the delete button or the Delete key.

On a phone, pinch to zoom and use two fingers to pan. Zooming in is worth the second it costs when the thing you are covering is small — a house number across the street, the name on a lanyard. Resize corners stay at least 44 pixels across at any zoom, so a thumb can still reach them. With a mouse the gestures are the same, and the pointer changes shape over a corner.

Keyboard

Give the image focus first, then a selected box responds to these:

← → ↑ ↓

Nudge the selected box by one pixel.

Shift + arrow

Move it by ten pixels.

Delete

Remove the selected box.

Esc

Deselect, leaving the box where it is.

Ctrl + Z

Undo. On a Mac, Cmd + Z.

Ctrl + Shift + Z

Redo the change you just undid.

Undo covers all of it: drawing, moving, resizing, deleting and mode changes. The history keeps your last hundred actions.

Which mode to use

Three modes, and only one of them deletes anything.

A solid bar fills the region with flat black. The pixels underneath are overwritten and gone from the exported file — there is nothing left in the image to work back from. That is why solid is what the tool starts with.

Pixelation splits the region into blocks and replaces each block with the average of the pixels inside it. That average is retained information: a very small version of the original, still sitting in the file. Blur does less again — three stacked box passes spread each pixel into its neighbours, which attenuates detail rather than removing it. Both are partially recoverable, and how much so depends on the block size and on how many pixels the face occupied to begin with.

So: pixelation or blur when the photo still has to read as a photograph of a real place, and a solid bar when what is underneath must not come back. Blur, pixelate, or black box works through the full argument and the cases where the difference matters more than people expect. You can switch the mode on a selected box at any point, or set every box at once.

What the detector was not looking for

BlazeFace finds faces. It finds nothing else, and it does not try to. Go over the frame once more for the things that identify someone just as well: a name badge, a house number beside a door, a screen with an inbox open on it, a delivery label, a license plate, a document face-up on a table, a reflection in a window. Draw a box over each one by hand.

There is no automatic plate detection here and no automatic text detection, and that is a stated limit rather than an unfinished feature — no reliable, permissively licensed model for either runs in a browser today. Drawing one box is not a hardship. If a car is the reason you came, blurring a license plate opens this same editor with a solid bar ready to go. For account numbers, addresses and anything else made of words, blacking out text is the same tool aimed at the same job.

Nothing here is saved

The photo, the boxes and the undo history live in this browser tab and nowhere else. There is no account, no draft, no autosave and no copy on a server, because after this page loads there is no server in the picture at all. Closing the tab loses the work. That is the deliberate cost of never uploading the image — so finish the save before you close it.