Screenshot to Text — Free, Fast & Private
Most people who go looking for this end up staring at an extension store listing. Somewhere between five and fifty of them promise to grab text off your screen, and the good ones genuinely work. The catch is the permission prompt that comes first, and what it asks for is not modest. Skip that entire decision. Everything on this page is a web page: take your screenshot however your computer already takes screenshots, drop the resulting file into the box above, and the recognition runs right here in the tab. There is nothing to install, nothing added to your toolbar, no account, no sign-in, and no permission dialog, because a web page you visited cannot see anything except the file you deliberately handed it. The engine is a self-hosted build of Tesseract 7 compiled to WebAssembly; it and a roughly 1.9 MB English model arrive from this site on your first visit and are then cached, so every screenshot after that starts reading immediately, even offline.
What that extension is actually asking for
To read text off any page, an extension needs to be able to see any page. So the install prompt says something close to: read and change all your data on all websites. Grant it, and the extension can observe your webmail, your bank dashboard, your admin panels, your internal tools, and every URL you open, for as long as it stays installed. Nothing about that is dishonest — it is the minimum an extension of that kind requires — but it is a very large standing permission to hand over in exchange for occasionally copying a sentence out of a picture.
The bigger issue is that the permission outlives your decision. Extensions update themselves silently, and the developer who wrote the one you trusted can sell it; there is a well-documented pattern of small, useful, well-reviewed extensions changing owners and quietly acquiring analytics, affiliate link rewriting, or worse, all pushed to existing users under the same name and the same rating. You reviewed version 1.4. You are now running version 2.9. Nobody asked you again. A web page has none of that leverage, which is why the honest answer to "which screenshot OCR extension should I install?" is often "none, if a tab will do."
Taking the shot: macOS
Press Cmd-Shift-4 and drag a box around whatever you want to read. The capture lands on your Desktop as a PNG, ready to drag straight into the box above. Two variants are worth knowing. Hit Cmd-Shift-4 and then tap Space, and the pointer becomes a camera that captures one whole window cleanly, including its shadow, without you having to trace the edges. And Cmd-Shift-5 opens the capture toolbar, where you can change the save location, set a timer, and choose between screen, window, and selection. If you only ever remember one, make it Cmd-Shift-4: it is the fastest path from "I need those words" to a file sitting on the Desktop.
Taking the shot: Windows
Press Win-Shift-S to open the Snipping Tool overlay and drag a rectangle. The snip goes to your clipboard, and a notification appears in the corner; click it and you can save the snip as a file, which is what you want here. If you would rather skip that step entirely, Win-PrtScn captures the whole screen straight to a file in Pictures\Screenshots with no dialog at all. Both routes give you a PNG you can drag into this page. Windows also lets you remap the physical PrtScn key to open the Snipping Tool overlay, which is worth ten seconds in Settings if you do this often.
Why screenshots are the easiest thing OCR ever gets
A screen capture has no camera between the text and the file. The characters were drawn by a font renderer at whatever resolution your display runs, then recorded exactly: perfect focus, perfectly horizontal baselines, even illumination across the entire frame, and contrast chosen by a designer who wanted the text to be legible. None of the things that wreck recognition on paper — shadow, glare, blur, perspective, curl — can occur. Whatever your screenshot tool saves, this page will take, along with the other formats it accepts: JPEG, PNG, WebP, AVIF, HEIC, BMP, GIF, and TIFF. That last pair matters more than it sounds, since a screenshot taken on an iPhone and sent to your laptop arrives as HEIC, and plenty of browser-based tools simply refuse it.
One thing does trip people up: capturing a region that is very small on screen. If the interface was zoomed out, the letters may be too few pixels tall to identify, and blowing the image up afterwards will not bring back detail that was never captured. Zoom in first, then snip.
The privacy argument, stated plainly
Consider what a screenshot usually contains. People capture the things software refuses to let them copy: a payment confirmation, a two-factor code, a medical portal, an internal dashboard, a conversation, a document rendered in a viewer that blocked selection. Nearly every browser-based OCR service handles these by posting the picture to a server, recognising it there, and sending text back, which means the most sensitive images on your machine are travelling to a company whose retention policy you have never read. Here the picture is decoded and analysed by code running in your browser, so no request carrying image data is ever made. You do not have to believe that on faith: open the network panel in your developer tools, run a screenshot through, and watch. After the first visit has cached the engine, you can also disconnect from the internet entirely and it will keep working — a thing that is simply not possible if a server is doing the reading.
The limits, before you find them yourself
Three honest caveats. The output is plain text, so a captured table loses its columns and a settings panel arrives as labels and values run together in reading order. The model reads printed and rendered characters, which covers everything on a screen but not handwriting, so a photo of a notebook is the wrong job for it. And this version speaks English only, with more languages coming; each one is another model file that has to be downloaded and cached, so they get added deliberately rather than promised vaguely. There is also no PDF input today. If your text is in a PDF, check whether you can already select it before reaching for recognition at all — a great many PDFs carry a text layer and need no OCR whatsoever. If you want the same engine framed around a different job, the general image-to-text page covers the wider case, and PNG to text goes deeper on why capture format matters.