Share the screenshot.Not the secrets.

What image redaction actually removes

Redaction is not decoration. A black bar, a pixelated block, a blur, and a crop all change an image, but only some of them remove the information underneath. The test is simple: after export, the covered detail should not be readable, recoverable, or inferable from what remains. If a name, email address, card number, ID number, address, date of birth, barcode, or account handle can still be read — or reconstructed from the pixels — the image is not redacted yet.

That distinction matters because screenshots are full of combinations. A name on its own may be public. The same name beside a postcode, an order number, and the last four digits of a card is a package that can unlock an account or support a payment. Cover what identifies a person, grants access, or proves a transaction, and treat any two of those details together as more sensitive than either one alone.

Why screenshots hide more than you think

A screenshot records the whole frame, not the part you meant to capture. Browser tabs, the URL bar, notification banners, a signed-in account chip, an avatar, a workspace name, a timestamp, and the file name in a window title all travel with the image. The detail you noticed first is rarely the only one worth covering.

Image files can also carry metadata: EXIF fields, GPS coordinates, camera and device details, and the original file name. Screenshots hold less of it than camera photos, but assume anything the recipient does not need should not leave with the file. This editor exports a newly encoded image from a canvas, so the original metadata stays behind.

How AI image redaction software works here

The detection in this image redaction software is OCR plus pattern checks, not a generative model guessing at faces. Tesseract reads the text already visible in the image inside a Web Worker, then the editor looks for emails, Luhn-valid card numbers, phone numbers, SSNs, Aadhaar numbers, PANs, GSTINs, IBANs, IP addresses, dates, address keywords, and short capitalised lines that look like names. Every match arrives as a suggestion you can keep or dismiss.

AI image redaction software is useful as a first pass, not a final authority. The scan can miss a face, a QR code, a barcode, a logo, or an unusual name, and it cannot know which details are sensitive in your context. Use it to stop hunting line by line, then review the frame yourself. All text mode covers every detected line when a busy screenshot makes individual judgement calls feel like a gamble.

Image redaction online, without the upload

Most image redaction online tools ask you to send the file to a server, wait for it to come back, and trust that the copy was deleted. This one is built the other way around. The browser decodes the image into a canvas in the current tab, keeps the pixels in memory, and draws every redaction locally. There is no upload endpoint, so there is no server copy to delete. The privacy page documents every request the site makes.

One request worth knowing about is smart detection. On first use, the page downloads Tesseract.js and an English language model. Your image is not part of that request. Manual redaction, undo, copy, and export work without it, and once the page has loaded you can keep editing offline.

Four styles, and the one to trust

Ink black draws an opaque, flattened shape over the original pixels. There is no gradient to reverse and no pattern to analyse, which is why it is the default for IDs, payment details, addresses, and anything that grants access to an account. Pixelate replaces an area with averaged colour blocks; it can look decisive while preserving the shape of large letters. Blur is the softest option and the weakest guarantee, because lightly blurred or pixelated text and faces can sometimes be partially reconstructed.

The emoji sticker draws an opaque plate before the emoji, so it covers the area instead of floating transparently over it. It suits a face in a casual group chat. Cropping is the strongest redaction of all when the surrounding context is not needed: a crop cannot be unblurred. For anything that identifies a person, moves money, or opens an account, reach for ink or a crop. Save the softer effects for context you would not mind losing.

What a good image redaction tool should do

The feature list matters less than the boundary around it. A tool worth using should work on the file locally, keep every box editable until you export, flatten the result into new pixels, and verify its own output before you download. It should also be honest about where detection stops instead of implying that a scan is a guarantee.

Those principles are the reason this editor works the way it does:

• Local by default: the file is decoded in the tab, with no upload endpoint.
• Editable until export: boxes move, resize, undo, and redo; the download is flattened.
• Verified output: a re-scan flags text that still reads as sensitive.
• Clean files: fresh PNG, JPG, or WebP exports without the original metadata, plus a size panel that measures the real download.
• No account, no watermark, no payment.

The useful question is not how many formats a tool lists. It is whether the file, the pixels, and the metadata all end up where you expect.

A short check before you send

Before the image leaves your hands, make one slow pass over the whole frame. These checks catch most of the expensive mistakes:

• Look at all four corners, the header, the tabs, and any notification.
• Use ink for identifiers and payment details; reserve blur and pixelate for context.
• Zoom in and confirm that no text peeks out from behind an edge.
• Run Verify, then open the exported file at full size.
• Ask whether a crop would be safer than a redaction.
• Send only the redacted version. Never follow it with the original.

Receipts, IDs, work dashboards, and chat screenshots each have their own guide. The short version is always the same: cover what identifies, verify what remains, and send the smallest image that still makes your point.