How to copy text out of a screenshot

Retyping an error message from an image is one of those small daily indignities that has been solved for years without most people noticing. Your operating system can probably already do this, for free, and you should try that first. This page covers the built-in routes honestly, and then the narrower set of situations where they are not the right answer.

On a Mac, the text is already selectable

Recent versions of macOS read text in images automatically through a feature called Live Text. Open a screenshot in Preview, Quick Look it with the space bar in Finder, or view it in Photos, and then drag across the words as if it were a document. The cursor changes to a text cursor over recognised text.

Select what you want and copy it with Command and C. There is no mode to enable and no button to find. If dragging only draws a selection rectangle rather than selecting words, look for the small Live Text button in the corner of the image, which toggles text selection on.

It handles multiple languages, it is fast, and it costs nothing. For a screenshot sitting on your desktop that you want one sentence out of, this is the answer and nothing else needs to be involved.

On Windows, use Snipping Tool or PowerToys

Windows 11 Snipping Tool includes text extraction. Open the screenshot in Snipping Tool and use the text actions button to pull the words out, copy all of the text at once, or copy a selected part of it. It can also mask things like email addresses and phone numbers before you copy, which is a thoughtful touch for a screenshot you are about to paste into a ticket.

Microsoft PowerToys, a free download from Microsoft, adds a tool called Text Extractor that works from anywhere. Press its shortcut, drag over any text on screen, and it is on your clipboard. Because it reads the screen directly, it works on text inside applications that will not let you select it, which is the case that annoys people most.

So when is a browser tool the right answer?

Being straight about this matters more than selling you something. If you are on your own Mac or your own Windows machine, taking text out of one screenshot, use the tools above. They are free, they are already installed, and they are good.

There are three situations where they do not help.

You cannot install anything. Managed work laptops routinely block PowerToys and everything like it, and an older Windows build may not have text actions in Snipping Tool. A page you open in a browser is not an installation and does not need approval from anyone.

You are not on your own machine. A borrowed computer, a shared workstation, a Chromebook, a machine in a lab. Nothing to set up and nothing left behind afterwards.

You want the text kept, not just copied. This is the real difference. The built-in tools give you the words once and then forget them. If the text is stored alongside the image, the screenshot itself becomes findable later by what it said, which is a different and larger problem than getting one sentence onto your clipboard today.

How it works here

Paste a screenshot into Screenshot Files and the recognition runs in the page, on your own machine. The engine is downloaded once and then works locally, so the image is not uploaded anywhere to be read. What gets stored is the recognised text, which is what makes the screenshot searchable afterwards.

That local-only detail is not a technicality. Screenshots are the most sensitive category of file most people produce without thinking: customer records, internal dashboards, private messages, whatever else was on screen. A tool that ships those to an outside vision service to make them readable has traded your data for convenience. Doing the reading in the browser avoids the trade.

Two honest limits. Recognition is currently English, so interface text in other languages will produce useful fragments like names, numbers and domains rather than a clean transcript. And reading an image uses your own processor for a second or two rather than someone else's, so a large batch is noticeably busier than doing nothing.

Getting a clean result

Recognition quality depends mostly on the input, and screenshots are unusually good input. They are flat, evenly lit, at a sensible resolution, and made of rendered text rather than printed text photographed at an angle. That is the easy case.

Two things still help. Capture at full resolution rather than scaling the window down first, because recognition works from pixels and a downscaled screenshot has thrown some away. And crop to the part you actually want before recognising, which removes the surrounding interface chrome that otherwise arrives mixed into your result.

Do not expect a perfect transcript from a screenshot of a photograph of a screen. That is a different and much harder problem, and no tool on this page solves it well.

Common questions

Does the image get uploaded to be read?

No. Recognition runs in your browser. Only the recognised text is stored, so that search has something to match against later.

Can it read handwriting?

Not reliably. The engine is built for rendered and printed text. Handwriting is a different problem and you should expect poor results.

Which languages work?

English. Text in other languages often still yields usable fragments, particularly names, numbers, URLs and error codes, but not a clean transcript. If you need multilingual recognition today, the macOS Live Text route above handles it better.

Why is the first screenshot slower than the rest?

The recognition engine is fetched the first time you use it and then reused. The first image waits for that download and every one after it does not.