Sharing a screenshot is routine: a bug report, a dashboard in a deck, a support ticket, a receipt. So is the small panic afterward, when you notice the customer email in the sidebar or the account number in the header. Redaction seems trivial — drag a blur over it and move on — but some published redactions do not actually hide what they were meant to hide.
Why redaction fails in practice
- Weak pixelation over text can be attacked. Pixelation is a deterministic transform. Given a known font and a small character set — digits, an email address — an attacker can render candidate strings, apply the same pixelation, and match the result. Published research and released tooling have demonstrated this against real screenshots.
- Light blur leaves recoverable structure. A small blur radius keeps enough edge information that sharpening or deconvolution can bring text back to a readable state.
- Cropping in some tools does not remove the pixels. Certain file formats and editors preserve the cropped-away region in the saved file. If you crop to hide something, verify the exported file, do not assume.
- Everything around the redaction leaks. URL bars, browser tab titles, notification banners, a name in the window title, an avatar, the shape of a chart. The box you drew is rarely the only clue in the frame.
- Metadata can survive an edit. Location, device, original timestamps, and sometimes a thumbnail of the unedited image can live in EXIF data.
Mosaic, blur, or opaque paint?
| Method | Best for | Watch out for |
|---|---|---|
| Opaque paint | Passwords, tokens, account and ID numbers, anything legally sensitive | Visually blunt; in visdraw, use the Paint brush at 100% opacity and cover beyond the content before exporting a flattened file |
| Mosaic | Faces, names and addresses in a demo, tidying a screenshot for a blog post | Block size must be large relative to the text. Small blocks over digits are weak. |
| Blur | Backgrounds, logos, non-critical context you want softened rather than erased | Small radii are reversible. If it is still legible when you squint, it is not enough. |
A practical heuristic for mosaic strength: each block should be at least as large as the height of the text you are hiding. If a line of text is 16 pixels tall, a 4-pixel mosaic is decoration, not redaction. Increase the block size until you cannot reliably infer the character count or shapes.
A redaction workflow that holds up
- 1
Screenshot the smallest useful region
Capture the panel, not the whole desktop. Every extra pixel is another thing to check. A window-only capture also avoids tab titles and notification banners.
- 2
Crop before you redact
In a flattened export, removing an area is generally safer than leaving it visible under a weak effect. Crop first so you only redact what genuinely has to stay in frame, then inspect the exported file to confirm that the removed area is gone.
- 3
Pick the method per item, not per image
Use 100%-opaque paint over the account number, mosaic over lower-risk customer names, and nothing over the parts that carry the point of the screenshot. Uniform treatment usually means over-redacting the harmless and under-redacting the critical.
- 4
Over-cover the edges
Extend the redaction a few pixels past the text on all sides. Ascenders, descenders and anti-aliased edges outside the box still describe the characters.
- 5
Export to a flattened raster format
PNG or JPG. The redaction becomes part of the pixel data with no separable layer, object or annotation for anyone to remove.
- 6
Reopen the exported file and inspect it
Open the actual file you are about to send, zoom to 300–400%, and read it as if you were trying to break it. This final check can reveal uncovered edges and missed details.
Pre-publish checklist
- Browser URL bar, tab titles and bookmarks bar
- Notification banners, badges and unread counts
- Usernames and avatars in headers, sidebars and comment threads
- Autocomplete dropdowns and recently-used lists
- Internal hostnames, IPs, ticket IDs and staging URLs
- Reflections in glossy surfaces if the screenshot is a photo of a screen
- File name —
invoice-acme-corp-2026.pngsays plenty on its own - EXIF metadata on camera photos: location, device, original timestamp
Do not forget the file itself
Screenshots generally carry little metadata; photographs carry a lot. A phone photo of a whiteboard can include GPS coordinates precise enough to identify the building. Re-exporting an image through an editor usually drops most of that, but the reliable move is to check rather than assume — and to rename the file to something neutral before you attach it.
Frequently asked questions
Can pixelated text really be recovered?
Under the right conditions, yes. When the font, layout, and character set are constrained — digits, an email address, a known UI — a candidate string can be rendered, pixelated identically, and compared. Public tooling exists for this. Use 100%-opaque paint and a flattened export for anything that truly matters.
How strong should a mosaic be?
Each block should be at least as large as the height of the text underneath it, and ideally larger. If you can still count the characters, increase the block size.
Is blur or mosaic better for faces?
Both can obscure a face when applied strongly, and mosaic is easier to judge because its strength is visible. Cover the whole head rather than just the face — hair, glasses, jewelry, and posture are all identifying.
Do I need to upload my screenshot to redact it online?
No. visdraw processes the image locally in the browser with WebGL and Canvas, so the editor does not transmit the screenshot to a server for image processing.
Which format should I export a redacted screenshot as?
PNG for UI screenshots — it is lossless and keeps text crisp. Avoid formats that preserve layers or annotations separately, because a separable layer can be removed. See the export format guide.
Written by
The visdraw team
Product & engineering