How to copy text from a screenshot or photo

Someone sends a screenshot of an error message, an address or a bank reference. You photograph a page of a book you want to quote, a receipt for an expense report, a serial number on the back of a router. Retyping is slow and error-prone. OCR, optical character recognition, turns the pixels back into text you can paste, search and translate.

This guide covers which option to reach for on each device, how to prepare an image so recognition works, and how to check the result before it ends up in a payment form or a contract.

Start with what is already on your device

For a single line of text on a phone, the built-in option is often the fastest. Be honest with yourself about the job before opening any tool.

iPhone and iPad

Recent iPhones recognise text in photos and screenshots automatically. Open the image in Photos, press and hold on a word, and drag the selection handles as you would in a document. The camera app can do the same with a live view. This works well for short snippets and mixed languages.

Android

Most Android phones have a lens feature in the camera or the gallery app that selects text in an image. The exact name and location depend on the phone maker, but it is usually one tap from the photo viewer.

macOS

Recent versions of macOS select text directly in Preview, Photos and Quick Look. Open the screenshot, move the cursor over the text until it turns into a text cursor, and select.

Windows

The Snipping Tool in Windows 11 has a text action that recognises text in a fresh capture and lets you copy it. On older Windows versions there is no built-in OCR in the everyday apps.

When to use a dedicated OCR tool instead

Built-in selection is designed for grabbing a phrase. It becomes awkward when you need a full page, a long chat log, text in a language your phone is not set up for, or a plain .txt file you can save. That is where a dedicated tool earns its place.

Browser OCR in three steps

1. Open Image to text and choose the language of the text: English, Russian, Polish, Turkish, Spanish, Portuguese, Indonesian, Vietnamese, German, French, Ukrainian or Italian.
2. Drop the image, or simply press Ctrl+V (Cmd+V on a Mac) right after taking a screenshot. Recognition starts on its own.
3. Copy the text, or download it as a .txt file named after the image.

When you choose any language other than English, English is recognised alongside it. Interface labels, product names, URLs and code are usually Latin even inside a Russian or Vietnamese screenshot, and the second model keeps them from being garbled. Changing the language after a run re-reads the same image, so there is no need to drop it again.

The engine is Tesseract, running inside your browser tab. The image is not uploaded anywhere, which makes the tool suitable for ID cards, bank statements and internal screenshots. The first run downloads the engine and language data; later runs start almost immediately. JPG, PNG, WebP and BMP images up to 30 MB are accepted, one image at a time.

Preparing the image

OCR accuracy depends far more on the input than on the engine. A few minutes of preparation saves more time than any amount of proofreading.

Crop to the text

Toolbars, wallpaper, fingers and the edge of the desk all give the engine shapes to misread. Crop to the area you need with Crop image. Its frame keeps a fixed proportion (the original, 1:1, 4:3, 3:2, 16:9 or 9:16), so pick the one closest to the shape of the text block and zoom to fit.

Make the letters big enough

Tesseract works best when capital letters are roughly 20 to 40 pixels tall, which is what a page scanned at about 300 DPI gives. Practical consequences:

• A screenshot at native resolution is usually fine. A screenshot that a messenger shrank on the way is often not; ask for the original file.
• Zoom in before taking the screenshot rather than enlarging the image afterwards. Enlarging adds pixels but not detail.
• For a printed page, fill the phone's frame with the page instead of photographing the whole desk.

Straight and evenly lit

• Keep lines horizontal. A tilt of a few degrees is tolerated; around 10 degrees errors climb fast, and sideways text mostly fails. Fix orientation with Rotate image.
• Photograph pages flat. The curve near a book's spine bends lines and causes errors there.
• Avoid glare and the shadow of your own phone. Side daylight beats a flash.
• Dark text on a light background is ideal. Light grey on white, or text printed over a photo, is the hardest case.

iPhone photos in HEIC

iPhones save photos as HEIC by default, and the OCR tool reads JPG, PNG, WebP and BMP. Convert first with HEIC to JPG. Screenshots on iPhone are already PNG and need no conversion.

PDFs are a different case

Before running OCR on a PDF, check whether it already contains text. Press Ctrl+F (Cmd+F) and search for a word you can see. If it is found, the PDF was produced from a document, and PDF to text extracts everything instantly and exactly, with no recognition errors.

If search finds nothing, the PDF is a scan: one picture per page. For one page, you can export it with PDF to JPG at 3x (216 dpi) and run it through Image to text. For a whole document, OCR PDF adds an invisible text layer under the page images, so the file becomes searchable and copyable while looking exactly as before. That one runs on the server, with one free run a day.

Checking the result

Never paste OCR output into anything that matters without reading it. Errors are predictable:

• Look-alike characters. 0 and O; 1, l and I; rn and m; 5 and S; 8 and B. In IBANs, order numbers, passwords and Wi-Fi keys, compare character by character with the image.
• Tables. Columns come out as words separated by spaces. Paste into a spreadsheet and split by spaces, or crop and recognise one column at a time.
• Multi-column layouts. Lines from neighbouring columns can interleave. Crop each column separately.
• Line breaks. OCR keeps the line breaks of the image. To turn the text back into paragraphs, remove the breaks in your editor before pasting into an email or document.
• Amounts. For invoices and contracts, read the figure from the image, not from the recognised text.

To count what you got or compare against another version, Word counter and Text diff are handy.

Troubleshooting

• The result is empty. The engine found nothing it considers text. Usually the letters are too small or blurred, or the wrong script is selected, for example English chosen for an image in Cyrillic.
• The text is full of nonsense characters. Background patterns or a photo behind the text are being read as letters. Crop tighter.
• Accented letters are wrong. Choose the actual language rather than English; each language model knows its own alphabet.
• Handwriting. Tesseract is built for printed text. Neat block capitals sometimes work; cursive does not. The live text features on phones handle handwriting better.
• A QR code instead of text. Use QR reader; OCR does not decode codes.

Checklist

1. For a phrase, try your device's built-in text selection first.
2. For a page or a private document, use Image to text and set the language.
3. Crop to the text, straighten it, and make sure letters are large and sharp.
4. Convert HEIC photos to JPG before recognition.
5. For PDFs, test with search: text PDFs go to PDF to text, scans to OCR PDF.
6. Proofread numbers and codes character by character before using them.