Pixelate an Image

How do you pixelate part of an image?

You drag a box over the spot that has to go. Then the next one. Then the one after that. Every box sits on the image at the same time, each with its own mode and its own strength, and the tool renders once at the end.

That is the difference from most pixelate tools on the web. There you cover one spot, save, reload the saved file and start again. A photo with four faces and two licence plates means six rounds and five intermediate saves, each one compressing the picture a little further.

After you draw a box you can move it, resize it from the corners and delete it on its own. When the preview looks right, you download.

Not every spot is a rectangle. Above the controls you can switch from Rectangle to Freehand and circle the spot instead. The start and the end of the line are joined for you, and only the area inside it is processed. Moving, resizing from the corners and deleting then work exactly as they do for a box.

Which mode should you use where?

Three modes per box, and the right one depends on what sits underneath.

Faces suit the mosaic. It averages each tile down to a single colour, and the result reads as deliberate censorship at a glance, which is usually the point. Blur is easier on the eye and smears edges instead of throwing them away, so at low strength more structure survives than the smooth surface suggests.

Text belongs under the solid bar. The bar replaces the pixels inside the box with one flat colour. Whatever stood there is no longer in the file, at any zoom level and under any filter. The peer-reviewed recommendation is blunt about it:

“We believe that a solid bar provides the best choice for all consequential redactions of text in an image.”

Hill, Zhou, Saul and Shacham, On the (In)effectiveness of Mosaicing and Blurring as Tools for Document Redaction, PoPETs 2016

Germany’s Lower Saxony data protection authority reaches the same place from the regulatory side and adds a second condition: draw the bar at 100 percent opacity, then export to a format such as JPG that carries no layers for the individual editing steps (Hinweise zum Schwärzen von Dokumenten). Both conditions hold here by construction. The bar is drawn fully opaque, not with a soft brush somebody could lift again with a contrast slider, and the download is a flat raster image with no layer left to undo. That is what redaction means when it is done properly.

What comes out of the download?

The image comes back at the resolution it went in at. A 4032 x 3024 phone photo stays 4032 x 3024. The on-screen preview may be scaled down so that dragging stays smooth; the export still runs against the full-size source. A PNG with transparent areas comes back as a PNG with transparent areas.

The file is written fresh, pixel by pixel, from the browser canvas. It is not a patched copy of the original. That has a consequence most pixelate tools never mention: whatever EXIF data the source carried has nowhere to go in the new file. GPS coordinates, capture time, camera model, serial number, the software field and the XMP block are gone, because they are never written in the first place.

The small preview image that JPEG files carry inside the EXIF block goes too. That is the part photo editors forget to regenerate: cover a face, leave the embedded thumbnail alone, and you ship the untouched original as a postage stamp inside the “redacted” file. It cannot happen here, because the old thumbnail is never carried over.

An embedded colour profile does not survive the trip either. For photos in a wide gamut the result can look a shade different afterwards. For a picture that already has boxes painted on it, that is rarely the problem. For a print master it is.

Which files does the tool accept?

One image per pass, up to 50 MB. PNG, JPEG, WebP, GIF (the first frame) and AVIF are accepted, the last one wherever the browser decodes it itself. HEIC is out of scope in this version: run iPhone photos through HEIC to JPG first, then come back here.

The file stays on your device throughout. There is no upload and no server holding it, and you can cut the network connection once the page has loaded without changing the result.

Very large images run into a hard limit that is not ours but the browser’s. Every canvas has a maximum edge length and a maximum area, and past that it hands back empty pixels instead of an error. When an image crosses that line, the tool says so and refuses to process it. A blank result would be the worse answer, because a white rectangle gives no sign that it was supposed to be a photo with a plate number in it.

Can pixelation be reversed?

It depends on what is under the mosaic. The answer splits three ways, and none of the three is a plain yes or no.

Text in a known typeface should be treated as recoverable. The 2016 study quoted above trained statistical language models to read pixelated text back out, and it worked even when the mosaic tile was larger than the font: the authors recovered 18-point text under a 24-point mosaic. Their own summary is that “mosaicing and blurring, despite their widespread usage, are not viable approaches for text redaction”. Saving the file as a JPEG afterwards does not fix it, which is a common and expensive misunderstanding.

A face under a mosaic is a different case. Nobody reconstructs the face. What “unpixelate” services produce is an invented face that happens to downscale to the same mosaic, and it can be a different person than the one in the photo. That does not make the picture anonymous: someone who already holds a list of candidate people and clean photos of each can match a pixelated face against that list. In a study from the same year, a network trained for the task picked the right person out of 530 candidates 57.56 percent of the time at a 16x16 mosaic, against 0.19 percent for guessing.

Under a solid bar there is nothing left. One side effect is worth knowing: the bar reveals its own width, and with it roughly the length of what it covers. For a postcode that is irrelevant. For one word out of a short list of possible words it is not.

Rule of thumb: text, numbers, ID documents, account details and anything on a screenshot get the bar. Faces, children in the background, door signs and street scenes can have the mosaic.

Frequently Asked Questions

The questions that usually come up before the first download.

How do you pixelate part of an image?

Drag a box over the part that has to go. Every box stays on the image at the same time, each with its own mode and strength, and the tool renders once at the end.

How do I blur a face in a photo online for free?

Drag a box over the face and pick blur or mosaic. Free, no account, no watermark, no daily cap. Several people means several boxes in the same pass.

How do I black out text in an image?

Use the solid bar, not mosaic or blur. The bar replaces the pixels underneath with one flat colour, so nothing is left in the file to recover.

Can pixelation be reversed?

For text yes, for faces partly, under a solid bar no. Pixelated text in a known typeface can be reconstructed, which is why text belongs under a bar.

How do I redact a JPEG?

Drag a bar over each sensitive area, then download. The export is a flat raster file with no layers to peel back, and JPEG compression is not a redaction step on its own.

Does my image get uploaded to a server?

No. All processing runs inside your browser tab. No upload, no server, no sign-up, no tracking.

What file sizes and formats does the tool accept?

PNG, JPEG, WebP, GIF (first frame) and AVIF, up to 50 MB, one image per pass. HEIC is not in this version.

Does the download keep the original resolution?

Yes, pixel for pixel. A 4032 x 3024 phone photo comes back as 4032 x 3024, because the export always runs against the full-size source.

Does the exported file still contain EXIF and GPS data?

No. The download is written fresh from the browser canvas, so GPS coordinates, capture time, camera model and the embedded EXIF thumbnail have nowhere to travel.

Which Image Tools Are Related?

Other tools in the kittokit ecosystem that fit this workflow:

• Remove metadata, lists the EXIF, GPS and XMP fields in a file before removing them, PDFs included.
• HEIC to JPG, converts iPhone photos into a format this tool accepts, locally and without an upload.
• Background remover, cuts the subject out when the whole background has to go rather than a few spots.