The quick answer
Extracting text from an image is done with OCR, optical character recognition: software looks at the pixels and returns the words as characters you can copy. You rarely need to install anything. Windows, macOS, iOS, and Android all have a built-in way to select text in a picture, and an online tool such as the OCRphoto image to text converter handles the files those miss.
- Use the clearest copy of the image you have: the original photo or screenshot, not a copy of a copy.
- Open it with your device's built-in text tool, or upload it to an online OCR tool.
- Select or copy the extracted words.
- Check names, numbers, and dates against the picture before you use them.
On Windows
In Windows 11, take a capture with the Snipping Tool (Windows + Shift + S) and press the Text actions button to select or copy the words in it. Microsoft PowerToys adds Text Extractor, a system-wide shortcut (Windows + Shift + T) that lets you drag a box over any part of the screen and copies the text inside it.
For an image file already on disk, OneNote has a Copy Text from Picture option in the right-click menu of an inserted image.
On a Mac, iPhone, or iPad
Apple calls it Live Text. On a Mac, open the image in Preview or Photos and move the pointer over the words: it turns into a text cursor, and you can select and copy like in a document. On an iPhone or iPad, open the picture in Photos, touch and hold a word, and drag the handles to select, or tap the Live Text button in the corner to show every recognised line.
Live Text also works on screenshots and in the Camera app, which is the quickest way to grab a line from a printed page without saving a photo at all.
On Android
Open the picture in Google Photos and tap Lens, then Text, and select the words you need. The same Google Lens view is in the camera on most Android phones, and Circle to Search lets you select text on anything currently on screen.
On a computer, Google Drive can do the same for an image file: right-click it, choose Open with, then Google Docs, and the document that opens contains the recognised text under the picture.
Online, when the built-in tools fall short
The built-in tools are fast and free, and they fail quietly. Small type, faint print, handwriting, and images that have been shrunk or recompressed on the way through a chat app are where they return half a line or nothing. They also only work on the device in front of you.
An online OCR tool reads the file itself, on any device. With OCRphoto you upload the image, get the words back as plain text in reading order, and copy them or save them as a .txt file; the uploaded file is deleted after it is read. There are pages for the common starting points: photo to text, screenshot to text, JPG to text, PNG to text, handwriting to text, and scanned PDF to text.
What decides whether the words come out right
OCR reads pixels, so it cannot recover detail that is not in the file. Three things matter most: the text has to be big enough in the image, sharp, and saved without heavy compression. For photos of paper, add even light and a camera held parallel to the page — how to take a photo that OCR can read covers that.
We measured the size and compression part on one invoice with twelve details that must come out exactly (amounts, an IBAN, a date, a postcode, an email address), using the model this site runs. Every detail survived at normal screenshot size, even as a quality-10 JPEG with one small slip. Shrunk to 342 pixels wide, the PNG still read all twelve, but the same image as a quality-30 JPEG came back with only four to seven right.
| Same invoice, saved as | 760 px wide | 342 px wide |
|---|---|---|
| PNG | 12, 12, 12 | 12, 12, 12 |
| JPEG quality 75 | 12 (at 60 and 95) | 11, 11, 11 |
| JPEG quality 30 | 12, 12, 12 | 4, 7, 6 |
| JPEG quality 10 | 11, 11, 11 | Unreadable |
Check the numbers before you trust them
The dangerous OCR mistake is not garbage, it is a clean result that is wrong. In the test above, the small quality-30 JPEG came back as a neat, well-formatted invoice with the total as 1,930.50 instead of 2,121.50 and two digits of the IBAN changed. Nothing in the output looked broken.
So check the parts that carry no context: amounts, dates, account and phone numbers, codes, email addresses, and URLs. Characters that look alike are the usual suspects — 0 and O, 1, l and I, rn and m, a comma and a full stop. For anything you will pay, send, or sign, compare against the image line by line.
Extracting words from a screenshot, a scan, or handwriting
Screenshots are the easiest input: computer-rendered text is sharp and evenly lit. Before running OCR on one, check whether the original page or app still lets you select the text; how to copy text from a screenshot on Mac or Windows walks through both.
A scanned PDF is a set of page pictures, so it needs OCR too; how to OCR a scanned PDF explains how to tell. Handwriting reads well when it is dark and evenly spaced, and needs a closer check when it is joined up or cramped — see how to convert handwriting to text.