Zellio.io

Image Redactor

Permanently black out, blur or pixelate parts of an image, on-device.

Processed locally — nothing is uploaded

About Image Redactor

The distinction worth knowing: a shape drawn over a picture in a document or slide is a cover, not a removal — whoever receives the file can select it and drag it aside. This rewrites the pixels themselves and re-encodes, so the covered detail is absent from the exported file rather than hidden inside it.

Covering something in a screenshot usually means dropping a filled shape on top of it in a document, a slide or a design tool. That hides the detail from view without removing it: the picture beneath the shape is still whole, and anyone who opens the file can move the shape aside. Newsrooms, law firms and government departments have all published documents this way and had the covered material read back to them.

This works the other way round. The area you drag over is rewritten in the image buffer itself — averaged into blocks, smeared, or painted flat — and the file you download is encoded from those rewritten pixels. There is no layer to peel off because there is no layer, and the detail that was there is not somewhere else in the file either.

The black box is the default, and it is the only one of the three that cannot be reversed by any algorithm: the area becomes a flat colour, so the result does not depend on what was underneath and there is nothing left to reconstruct from. Pixelating and blurring are a different kind of thing. A mosaic resembles the original because it still encodes its broad shape, and published research — Depix being the best known — recovers pixelated text from screenshots by rendering candidate strings, averaging them the same way and matching the blocks. Blur is a linear operation and can be substantially undone. Treat both as cosmetic: fine for keeping a reader's eye off something, wrong for anything you would mind being read.

Everything happens in the tab. The image is decoded by the browser, edited in memory, and encoded again when you press download — it is never sent anywhere, which also means there is no copy on a server to think about afterwards. A useful side effect: because the download is a fresh encode rather than a modified original, EXIF fields travelling with the source photo, including GPS coordinates and the camera's serial number, do not come with it.

Learn how this works

How to Edit Images in the Browser Without Uploading ThemCompress, watermark and strip metadata from photos entirely on your device — why client-side image editing is faster and more private.8 min read

Frequently asked questions

Can the covered part be recovered from the downloaded file?

With the black box, no — the area is set to a flat colour, so the output carries no information about what was there and no algorithm can recover it. With pixelate or blur, treat the answer as yes: both leave the original’s broad structure in the image, which is what makes a mosaic look like the thing it covers, and there are published methods that read pixelated text back. Separately, and unlike a shape dragged over a picture in a document, the download here is encoded fresh from the edited pixels, so nothing is hiding underneath a layer and no leftover data trails the end of the file.

Should I use blur or pixelate?

For anything sensitive, neither — use the black box. Pixelate and blur both survive as a low-resolution version of the original, and recovering text from a mosaic is a solved problem. Between the two, pixelate discards more than blur, and a bigger block discards more than a small one, so pixelate at high strength is the better of a bad pair. They earn their place on faces in a photo meant for people rather than as evidence, where the point is that the subject is not identifiable at a glance.

Does the exported image still contain the camera and location metadata?

No. The download is encoded fresh from the edited pixels rather than patched into the original file, so EXIF data — including GPS position, timestamp and camera identifiers — is not carried over.

How does "Find faces" work, and does the photo go anywhere?

It runs BlazeFace, a small face detector, on your own machine. Pressing the button downloads about 450 KB of model weights from this site the first time — nothing before that, which is why it is a button rather than something that happens when the page loads — and the detection itself runs in the tab against your graphics hardware. The photo is never sent anywhere, and the model comes from here rather than a third-party CDN. What comes back is ordinary ellipses: move them, resize them, undo them, add your own. A detector finds most faces in a clear photo and misses profiles, partial faces and anyone small in the background, so it is a first pass rather than an answer.

PNG or JPEG?

PNG if the image is a screenshot or has text and sharp edges: it is lossless, so nothing outside the covered areas changes. JPEG produces a much smaller file for photographs, at the cost of re-compressing the whole picture, which very slightly alters parts you did not touch.

Is my image uploaded anywhere?

No. Everything runs locally in your browser using standard web APIs — your text, files and inputs are never uploaded to a server, so the tool works even offline once the page has loaded.

Is it free?

Completely. There is no sign-up, no account, no watermark and no usage limit. The tool is supported by unobtrusive ads, not by selling or processing your data.

Pro Tips

  • →Zoom the browser in before dragging if the area is small — the region is stored as a proportion of the image, so it stays exactly where you put it at any zoom.
  • →Cover a little more than looks necessary. Partial characters at the edge are often enough to reconstruct a name or a number.
  • →Use the ellipse for a face and freehand for anything awkward — a rectangle round a head either leaves the jaw showing or takes half the background with it.
  • →Find faces does the first pass on a group photo, deliberately covering more than the detector outlined so the hairline and jaw go too. Check every one and drag over whoever it missed — profiles and background faces are exactly what it drops.
  • →Check the result at full size before sending it. The preview is the same pixel data as the download, so what you see is what the file contains.
  • →For a screenshot containing text, prefer the black box: there is no residual pattern to analyse at all.

Common Use Cases

  • •Removing names, account numbers or email addresses from a screenshot before pasting it into a bug report or a public issue tracker.
  • •Hiding faces or number plates in a photo before posting it.
  • •Clearing customer data out of a dashboard screenshot for a slide deck or a case study.
  • •Stripping identifying detail from a document photograph shared with support.

Related tools

Watermark AdderOverlay custom text or a logo onto any image, then download it.Meme GeneratorAdd bold top and bottom text to classic or custom images.Image CompressorShrink PNG & JPEG files to a target size — one at a time or in batches.EXIF Viewer & RemoverInspect photo metadata and strip EXIF/GPS data before sharing.