How to Copy Text from an Image in About 10 Seconds

You have a picture with words in it and you need those words in your clipboard. Not a survey of every tool that exists — just the shortest path from image to paste, and a plan for when the result comes back mangled.

(If you want the full comparison of phone, desktop and browser options, that is how to extract text from an image. If you need the output as a spreadsheet or a formatted document, that is how to convert an image to text.)

The 10-second version

1. Open copy text from an image.
2. Drag the picture onto the page — or press Ctrl + V / ⌘ + V if it is already in your clipboard.
3. Wait a second or two while it is recognised.
4. Hit Copy.

That is the whole thing. No account, no upload, no install; the first visit downloads the recognition model, after which the page works even with the network off.

Copy one part, not the whole picture

This is the step most people skip, and it is the one that saves the most cleanup. A photo of a restaurant menu contains nine dishes you do not care about. A screenshot of a dashboard contains the whole sidebar.

Once the preview appears, drag a box around just the region you want, or click a detected block to snap to it. You get that region and nothing else, so there is no deleting surrounding junk in a text editor afterwards. For a picture of a page where you need a single quoted paragraph, this turns a two-minute cleanup into a single drag.

When the copy comes back wrong

Bad output almost always traces back to one of four things, in this order of frequency.

The text was too small. This is the big one. Optical character recognition needs roughly 20 pixels of height per lowercase letter to be dependable. A photo of a whole page taken from across the desk gives each letter maybe eight. Recapture closer, or zoom the page to 150% before screenshotting. Enlarging a blurry image afterwards adds no information — the detail was never recorded.

The shot was at an angle. A slight tilt is fine. A 30-degree perspective shot of a page is not: letters at the far edge are compressed into shapes the model was never trained on. Shoot documents from directly above.

The contrast was poor. Faded thermal receipts, grey-on-grey interface text, heavy watermarks. Models are trained overwhelmingly on dark text on a light background, so anything that flattens that gap costs accuracy.

There was glare. A phone photo of a glossy magazine page or a laminated sign frequently has a bright blown-out patch across a line. Nothing can be recovered from pure white pixels. Move, tilt the page, or block the light source.

Seven ways to get better OCR results goes through all of this in more depth, including language selection and cropping.

Characters that get confused

Even on a clean image, some pairs are genuinely ambiguous in many fonts: l and 1 and I, 0 and O, rn and m, 5 and S. This matters for prose far less than it matters for anything you are going to paste somewhere strict — a serial number, an error code, a password, a line of code.

The result panel stays editable for exactly this reason. Read the output once against the image before pasting it somewhere that will not tolerate a typo. Fixing one character in place is quicker than discovering the problem three steps later.

Tables

If what you are copying is a grid — a price list, a results table, a schedule — copying it as plain text gives you a run of words with the column structure destroyed. Grid-like regions get reconstructed as actual tables instead, which you can take as Excel, Markdown or CSV and paste straight into a spreadsheet.

Why this does not upload your picture

The recognition model is compiled to WebAssembly and downloaded to your browser, so the image is processed on your own machine rather than on a server. For a photo of a menu that hardly matters. For a screenshot of a bank statement, an ID document or an internal dashboard, it is the difference between a private operation and handing a copy to a company you know nothing about. The longer argument is here.

If it is not really an image

Two neighbouring cases worth knowing:

• The text is in a PDF — use copy text from a PDF, which reads the text layer directly when there is one and only falls back to OCR for scanned pages.
• The text is on a web page that blocks selection — screenshot it and OCR the screenshot. Full guide here.