Pixelate Image

Reduce a rectangle — or the whole picture — to flat blocks of averaged colour, without the file leaving this tab.

Runs on your deviceImagesWorkingNo account needed

What this tool does

Pixelate Image lays a grid of squares over the area you name, averages the red, green, blue and alpha of every pixel inside each square, and repaints that square as one flat colour. The averaged values are written straight into the pixel data before the file is encoded, so the picture you save holds block colours and nothing underneath them. That is the difference between this and drawing a grey box in an editor over a copy that still carries the original layer.

Two modes decide what gets covered. The whole image turns the entire photograph into a mosaic; one rectangle asks for a left offset, a top offset, a width and a height in pixels, and averages nothing outside it. Those outside pixels keep their original colours, but they are not passed through byte for byte: the whole picture is re-encoded on the way out, so choosing a lossy format recompresses the part you did not cover along with the part you did. Block size runs from 2 to 128 px and starts at 16. The grid is anchored at the rectangle's top-left corner, so squares that run off the far edge are averaged over the fewer pixels they actually contain, and a rectangle stretching past the image is clipped to the part that exists rather than refused. Queue several photographs and each is processed in turn under the same rectangle and the same block size.

The honest limit is what a mosaic is. It reduces information rather than deleting the subject: every block still reports the average colour of whatever sat beneath it, and that average is a measurement someone can work backwards from. Where the covered content comes from a small set of possibilities — a registration plate, a six-digit reference, a short word in a familiar typeface — a candidate can be rendered, pixelated with the same block size and compared against your blocks until one matches. Small blocks make this easy, and below roughly 8 px a face is often recognisable without any of that effort. The defence is size: push the block size up until the whole region collapses into one or two flat squares, because an average taken over that much area carries very little back. Failing that, crop the picture so the thing is not in the file, which is only an option when it sits near an edge. Note too that only the copy you download is altered; the original on your disk is untouched, so tidying that up is your own step.

None of this involves a server: the photograph is decoded, averaged and written out again inside this tab, which is also why an aeroplane-mode browser runs it fine. The re-encode at the end has costs worth knowing. Camera, date and GPS tags are not carried into the new file. GIF, BMP and SVG inputs cannot be written back in their own format, so with the format left alone they arrive as PNG — and an animated GIF gives up only its first frame on the way in. Choosing JPG flattens any transparency onto white, and this tool has no setting to pick a different fill colour. There is also no preview: the result line names the rectangle it used and how many blocks it produced, but seeing whether that rectangle actually landed on the face means opening the saved file.

How to use it

  1. Add one photograph or several. The decoder handles JPG, PNG, WebP, GIF, BMP and SVG, and AVIF where the browser supports it. A HEIC, TIFF or camera RAW file is turned away instead, with a line explaining that no browser carries a decoder for it.
  2. Say what the mosaic covers. Whole-image mode mosaics everything. One rectangle is the mode for censoring a single thing, and takes a left offset, a top offset, a width and a height, all measured in pixels from the top-left corner of the full-size image.
  3. Set the block size. Between 2 and 128 px, 16 by default. Bigger squares average over more pixels and throw away more of the shape underneath; small ones leave enough structure that faces and digits survive.
  4. Decide what it is written back as. The output format is either whatever came in or one of JPG, PNG, WebP and AVIF, chosen from the list. Quality is a 10-to-100 dial parked at 88, read by the three lossy formats and ignored by PNG.
  5. Save it, then open it. Check the saved file rather than trusting the coordinates. A rectangle a few pixels short leaves a readable sliver along one edge, and that sliver is the whole point of the exercise.

Processing & privacy

Runs on your device. Your file is processed directly in this browser.

Common use cases

A photo with a number plate in the background

Cover the plate before the picture goes on a listing or a forum, so the vehicle it belongs to is not a lookup away for anyone who saves the image.

A screenshot showing an account or order number

Mosaic the identifier and leave the rest of the interface legible, which is what makes the screenshot worth sending in the first place.

A crowd shot published without everyone's agreement

Faces that happened to be in frame can be blocked out one at a time while the subject of the photograph stays sharp.

A mosaic used deliberately

Whole-image mode at a large block size turns a photograph into a coarse grid of colour — useful as a placeholder thumbnail, a spoiler cover or a stylised background.

Questions

Where does the block averaging actually happen?
In this tab, and nowhere else. Decoding, the block averaging and the re-encode are all canvas work done on your own machine. Cut the network and the tool behaves exactly as it does now.
Can pixelation be undone?
Not from the file itself — the blocks are all it contains, and there is no hidden original beneath them. It is not proof against reconstruction, though: the block averages are evidence, and where the covered content is drawn from a small set, such as a plate or a short number, guesses can be pixelated the same way and compared until one lines up. Where none of it may survive, raise the block size until the region is one or two flat squares, or take the thing out of the frame with Crop Image if it happens to sit near an edge.
What block size is enough?
It depends on how large the thing is in the frame. A face two hundred pixels across needs blocks in the tens of pixels before its features stop reading; the same 16 px default over a face that fills the frame barely disturbs it. Judge it against the saved file rather than the number.
Can I cover two faces in one pass?
No. One rectangle is applied per run. Save the result, add that file back as the input and pixelate the second area — or switch to whole-image mode if everything in the picture is sensitive.
My mosaic came out in the wrong spot — why?
The four values are pixels measured from the top-left corner of the image at its real size, not the size it appears on screen. An origin outside the picture is pulled back inside rather than rejected, and a width or height that overruns an edge is trimmed, so a wrong number produces a misplaced mosaic instead of an error.
Is EXIF and GPS data kept?
No. The file is written out fresh from the canvas, so no camera, timestamp or location tags survive. Read them with Image Metadata Viewer first if you need to know what the original carried.
What comes back if I feed it an animated GIF?
One still picture. The decoder supplies the first frame and nothing else, and since GIF cannot be written from the canvas the result arrives as a PNG unless you name a different output format.
Should I use this or Blur Image?
Both destroy detail in place. Blocks read as an obvious, deliberate cover; a blur looks softer and, at a low strength, can leave more of the shape intact. Pick whichever you would rather the viewer see, and size it generously either way.

Related tools