Paste or drop a screenshot, mark it up with arrows, boxes, text and numbered steps, then pixelate or redact anything sensitive. Everything stays editable until you export — and nothing ever leaves your browser.
PNG, JPG or WebP — a screenshot works great
Drag on the image to draw. Switch to Select to move, resize, recolour or delete anything you have already added — nothing is baked in until you export.
🔒 Everything runs on an HTML canvas in your browser — nothing is uploaded.
Point at exactly what matters with a colored arrow or highlight box before sending a screenshot to a coworker.
The exported PNG contains only the averaged blocks — not a translucent layer someone could peel off. For anything that truly must not leak, the Redact tool covers it completely.
Copy the edited image directly to your clipboard and paste it into Slack, email or a doc — no file needed.
Every arrow, box, label and blur stays a separate object for as long as the image is open. Switch to Select and you can pick any of them up, drag it somewhere else, pull a corner to resize it, change its colour, or delete it — including annotations you added twenty steps ago. The picture you see is redrawn from the untouched original each time, which is also why undo can go back sixty steps without storing sixty copies of the image.
The alternative, and what most quick browser annotators do, is to paint each shape straight onto the canvas as you draw it. It is far simpler to build, and it means an arrow is permanent the moment you release the mouse: place one badly and your only option is to undo everything drawn after it. Keeping the annotations as objects costs a little more work per frame and removes that whole category of frustration.
Blur is the exception that has to work both ways. While you are editing, each blurred region is recomputed from the original pixels every frame — that is what lets you move a blur box you dropped in the wrong place, and it means sliding one aside brings back what was under it. On export the canvas is rendered once and handed to the PNG encoder, so the file you actually share contains nothing but the averaged blocks. The original never reaches the file, and the in-memory copy goes away when you load another image or close the tab.
The purpose of an annotated screenshot is to direct attention, and most annotations fail by adding rather than removing. Three marks on an image are read; twelve are scanned and abandoned. If several things need pointing out, a numbered sequence with a caption list works far better than a cluster of arrows, because it imposes an order.
Contrast is what makes a mark visible. Red is conventional and works well over most interface screenshots, which are predominantly light and low-saturation, but disappears over a red error state — precisely the screenshot most likely to need annotating. Bright green or magenta are safer defaults over unpredictable content.
A dimmed overlay with the relevant region left at full brightness is the strongest technique available and is underused. It directs the eye immediately, needs no arrow, and unambiguously communicates that everything else is context rather than subject.
Blurring is not redaction. A Gaussian blur is a reversible mathematical operation in principle, and in practice text recovered from a moderate blur is often readable by deconvolution. Pixelation is worse, because it is a deterministic average over a fixed grid — for a known font and a limited character set, the original text can be recovered by brute force, and this has been demonstrated repeatedly against real redacted documents.
The reliable method is to draw an opaque solid block. Not a semi-transparent one, not a heavy blur — an opaque rectangle in a flat colour, applied before the image is exported. If a tool draws it as an editable layer, flatten the result so the underlying pixels are actually replaced rather than merely covered.
Check what else is in frame while you are at it. Screenshots routinely leak more than their subject: browser tabs disclosing other work, bookmark bars, notification banners, email subject lines, the account name in a title bar, a file path containing a real name, and the URL itself. Cropping tightly to the region of interest removes most of this at a stroke.
Capture at the display's native resolution rather than scaling afterwards. A screenshot from a high-density display downscaled to half size loses the crispness that made it legible; one captured small and enlarged is worse still, because interface text does not survive interpolation.
Export as PNG. Screenshots are flat colour with hard text edges, exactly the content JPEG degrades most, and the mosquito noise around glyph edges is immediately noticeable. PNG is lossless and usually smaller for this content anyway, because large flat regions compress extremely well.
Crop tightly. A screenshot of a whole 27-inch desktop to show one dialogue box wastes the reader's attention and forces them to hunt. Include enough surrounding interface to establish where you are and nothing more. If the image will be embedded in a document, the PNG optimiser reduces the file without touching a pixel.
Removed. The exported PNG is rendered with each blurred region replaced by its averaged blocks, so the file you share contains no trace of the original detail there — it is not a layer anyone can peel off. While you still have the image open the original is kept in memory, which is what lets you move or resize a blur you placed badly; that copy never reaches the exported file and is discarded when you load another image or close the tab.
Yes — copy a screenshot to your clipboard and press Ctrl/Cmd+V anywhere on this page to load it instantly.
No — loading, editing and exporting all happen on an HTML canvas in your browser. Nothing is sent to a server.
Not always. Recovering text from pixelated screenshots has been demonstrated repeatedly, because a short string drawn from a predictable alphabet — a card number, a postcode, a username — leaves enough signal in the block averages to be guessed at. Coarser blocks help, which is what the size slider does. But when the content genuinely must not leak, use the Redact tool instead: it paints a flat opaque block with nothing of the original underneath it in the exported file.
PNG. Screenshots are flat colour with sharp text edges — the content JPEG handles worst — and PNG is usually smaller for this material as well as lossless.