Convert Image to Text

How it works

1. Choose up to 10 JPG, PNG or WebP images, drop them in, or paste a screenshot with Ctrl+V.
2. Put the images in the order you want, rotate any that are sideways, and remove the ones you do not need.
3. Pick the language of the text and select Extract text. The first read loads the engine, so it takes longer; later reads start sooner.
4. Check the text against the image and correct any mistakes in the box. Copy it, download a .txt file, or send it to the Word Counter, for one image or for all of them combined. The text is passed in this browser tab only: it is not put in the address, saved or uploaded.

Options explained

Language

English, Arabic or Urdu. The language file is downloaded only for the language you choose. Choose the language that matches the text: a wrong language gives wrong letters even on a sharp image. One language per read; images that mix languages are not supported.

Enlarge before reading

Automatic, the default, enlarges only very small images (under 0.1 megapixel) to twice their size. Always 2× enlarges every image and Never leaves them alone. On our test images, text about 8 pixels tall went from mostly wrong to almost right with Automatic, but Always 2× made larger Arabic letters worse, so use it only to retry a photo whose text is small.

Shadows or uneven lighting

Off by default. It makes the engine judge dark and light locally instead of across the whole picture, which helps a photo where one side is in shadow. On our test images it fixed a shaded page, but it made faint, low-contrast and noisy images worse, so turn it on only when shadows are the problem.

Rotate

Turn an image left or right in 90° steps. Phone photos that store their rotation in the file are straightened for you. A page that is only slightly crooked is read without rotating; a sideways or upside-down page is not, and reads as nonsense or nothing until you rotate it.

How the result is worked out

• Your browser decodes each image, applies your rotation, and scales it: very small images are enlarged unless you turn that off, and images over 16 megapixels are scaled down so reading stays quick. When no rotation or scaling is needed, your file is handed to the engine as it is.
• The Tesseract engine, running as WebAssembly in a background worker, finds the lines of text and reads the characters using the language data you chose. One worker reads the images one after another, and Stop ends the work at once.
• The engine also reports a confidence score from 0 to 100 for each image. That score is the engine's own estimate and is not the share of words that are correct. Below 60, the page asks you to check the text carefully.
• Line breaks follow the lines in the picture. Nothing is spell-checked, corrected or rewritten, and your edits are never replaced without asking.

Example

A sharp screenshot of a paragraph, with English chosen, comes back as plain text with the same words and line breaks. You then fix any slips, for example a capital O read as a zero or the letters rn read as m, and copy the text into your notes or into the Word Counter.

Tips for better results

• Use a sharp, straight-on image with the text filling most of the frame. In our tests, capital letters about 10 pixels tall read correctly on clean text and about 6 pixels tall did not; larger is safer, especially for photos.
• Crop away borders, pictures and anything that is not text before you add the image.
• Choose the language that matches the text. A wrong language gives wrong letters even on a sharp image.
• If the picture is sideways or upside down, rotate it before reading. Photograph pages in even light and avoid shadows across the text.
• Read the result against the image. Common slips are 0 and O, 1 and l, and rn read as m.

Limitations

• OCR is not exact. Accuracy depends on how sharp, large, dark and straight the text is, and on the font, the language and the layout. The tool does not understand or correct what it reads, so check the result before you use it for anything that matters.
• Handwriting is not supported and is usually read badly or not at all. Decorative fonts, text on busy pictures and very small or blurry text also fail.
• Tables, columns and forms are read as lines of text. The layout is not rebuilt, and the order may need correcting.
• Only JPG, PNG and WebP images are supported. HEIC, GIF, TIFF, BMP, RAW and PDF files are not. A PDF cannot be read by this tool yet.
• Arabic and Urdu were tested only on clear printed text in common system fonts. Urdu in the Nastaliq style is likely to read poorly.
• Limits: 10 images at a time, 15 MB for each image, 40 MB in total, and 40 megapixels for each image. Very old or low-memory devices may run out of memory on large images.

Guides that go with this tool

• How to Get Accurate Text From a Photo or Screenshot

Frequently asked questions

Related tools

• Word Counter
• Bibliography Sorter
• Words to Pages
• Reading Time