Blur Image

Soften a whole picture, or smear one rectangle over a face, a number plate or an address.

Runs on your deviceImagesWorkingNo account needed

What this tool does

Blur Image applies a Gaussian blur to the pixels of your picture. The maths is the standard three-pass box approximation: three successive box averages, sized so that together they approximate a Gaussian of the strength you asked for. Each pass carries a running sum along the row or column, so the work scales with the number of pixels and not with the strength — moving the slider up costs nothing extra, where a true Gaussian convolution would get slower with every step. The strength slider is the Gaussian sigma, and detail is smeared over roughly three times that many pixels — around 8 softens a face, around 25 hides one.

Two details of the implementation are worth naming. Colours are premultiplied by alpha before the passes run, so a cut-out PNG does not develop a dark halo where its transparent edge meets the subject. And a flat area of colour comes back the same colour, rather than fading towards the edges of the frame the way a naive blur does when it runs out of pixels to average.

You choose between softening the whole image and blurring one rectangle. The rectangle is typed in as four numbers — left, top, width and height, in pixels from the top-left corner — because there is no selection box to drag over a preview. A rectangle that hangs off the edge is trimmed to what exists rather than refused, and the summary afterwards states the size and position actually used, so a second run can correct a guess. That rectangle is lifted out, blurred as an image in its own right and pasted back, which has one visible consequence: its border stays perfectly sharp. The blurred patch reads as a box, and nothing from outside the rectangle bleeds into it.

No server is involved: the picture is decoded, blurred and written out again inside this tab. The price of that last step is a fresh encode. Nothing from the original file's header is carried across, so EXIF, the capture date and any GPS tag are absent from the result; an animated GIF gives up only its first frame; and GIF, BMP and SVG inputs come back as PNG, because the canvas can write only JPG, PNG, WebP and AVIF. Those four are also what you can force the output to be. Pick JPG and transparent pixels are filled with white, with no setting here to choose a different colour. A file the browser has no decoder for — HEIC, TIFF, a camera RAW — is refused with a message that names the reason and what to do instead, rather than failing without one.

How to use it

  1. Add your image, or a batch of them. JPG, PNG, WebP, GIF, BMP and SVG are decoded, and AVIF on recent browsers. Each file is blurred in turn under identical settings — including the same rectangle, if you are blurring one.
  2. Blur the whole image, or one rectangle. Whole-image suits backgrounds and cover art. The rectangle option takes a left offset, a top offset, a width and a height in pixels, and is the one to use for hiding a face or a house number.
  3. Set the blur strength. The slider runs from 1 to 100 and is the Gaussian sigma, not a pixel radius. Low values soften; anything that has to be unreadable belongs at the upper end.
  4. Pick the format and quality. Stay with the input's format, or force JPG, PNG, WebP or AVIF. The quality dial covers 10 to 100 and sits at 88 to begin with; it reaches JPG, WebP and AVIF output and does nothing at all for PNG.

Processing & privacy

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

Common use cases

Hiding a face before you publish a photo

A bystander in a street shot or a colleague in an office picture can be covered by one rectangle, leaving the rest of the frame at full detail.

Covering an address or an account number in a screenshot

Support tickets and bug reports are full of screenshots carrying details nobody needed to send. Blur the line, keep the rest legible.

Obscuring a number plate in a listing

A car for sale still needs to look like the car. Blurring the plate leaves the bodywork, the wheels and the condition perfectly readable.

Making a background for text

Softening a whole photograph drops its contrast and detail, so a headline or a logo laid over the top stops competing with the picture behind it.

Batch-softening a set of images

Queue a folder of shots and the same strength is applied to each, so a gallery of backgrounds looks consistent instead of hand-tuned one at a time.

Questions

Is my photo uploaded?
No. Decoding, the blur passes and the re-encode all run in this browser tab on the canvas, which is why the tool still works with no network connection at all.
Is a blur enough to hide something sensitive?
Treat it as concealment, not deletion. A blur is a reversible-in-principle operation: a gentle one can be partly undone by deconvolution, and where there are few possible answers — a six-digit reference, a number plate — someone can blur candidates and compare the results against yours. A high strength destroys far more than a light smudge, and cutting the area out with Crop Image removes the question entirely.
Can I drag a box over the image instead of typing coordinates?
No. The region is described by four pixel values, and the result names the rectangle it actually blurred. Read the coordinates in an image viewer first, or blur wide and check the output before you send it on.
Why does the blurred patch have a hard edge?
Because the rectangle is treated as a separate small image, blurred and pasted back. There is no feathering along its border and no setting for it, so the blurred area is visibly a box.
What does the blur strength number mean?
It is the sigma of the Gaussian, so the visible spread is wider than the number suggests — detail travels roughly three times that many pixels. Doubling the figure does considerably more than double the softening.
What happens to a transparent PNG?
Transparency is preserved for PNG, WebP and AVIF output, and the alpha channel is blurred along with the colours, so a cut-out subject's edge softens too. JPG has no transparency, so those pixels come out white.
Is camera metadata kept?
No. Writing the image back through the canvas produces a file with no EXIF block, so the camera model, timestamp and any GPS coordinates are dropped as a side effect. Read them with Image Metadata Viewer beforehand if you need them.
Can I blur an animated GIF?
Only its first frame, which is all the browser hands over as a still image. The result is saved as a PNG, since the canvas has no way to write a GIF back.

Related tools