Image to text (OCR):
copy text from a photo

How to convert an image to text

1. Pick the Language of the text in the settings panel first. The right choice makes the biggest difference to accuracy.
2. Click Choose files or drop your pictures onto the page. JPG, PNG, WEBP, HEIC, AVIF, BMP, GIF and TIFF are accepted.
3. Wait while the text is read. The first time you use a language, the text reader downloads, so this takes a little longer.
4. Click Save to download the text, or Download all (.zip) for several pictures.

What the settings do

• Language. English, Urdu + English, Hindi + English, Arabic + English, Spanish, Portuguese or Indonesian. The combined settings read both scripts, which suits forms and notices that mix them. Pick the wrong language and the words come out garbled.
• Layout. Auto (finds columns) works out blocks and columns on the page, as in a letter or a magazine. Line by line (receipts, tables) reads the picture as one block, line after line, and keeps the spaces between words, so an item and its price stay on the same line.
• Save as. Text file (.txt) gives plain text. Word document (.docx) puts each paragraph in Word with its line breaks, and Arabic and Urdu paragraphs are set right to left.

What happens to your picture

Optical character recognition (OCR) looks at the shapes of the letters, not at any text stored in the file. Before reading, small pictures are enlarged so the letters are big enough, see-through areas are filled with white, and a local contrast step copes with shadows and tinted paper.

In Auto mode, a tilted or blurry photo can confuse the column detection. When very little is read, the picture is read again as one block and the better result is kept.

A multi-page TIFF, as scanners and fax software often write, is read page by page. The text file marks each page, and the Word document starts each one on a new page.

The text reader and the language data are downloaded once, about 1 MB for the engine plus 1 to 3 MB per language, and kept by your browser. After that, the same language works without a connection.

Limits worth knowing

• Printed text only. Handwriting isn’t read reliably; neat block letters sometimes work, joined-up writing usually doesn’t.
• Sideways or upside-down text isn’t turned the right way. Rotate the photo in your phone’s gallery first.
• You get the words, not the design. Fonts, bold, colours and table borders aren’t kept, and columns become separate blocks of text.
• Small, blurry or low-contrast pictures give mistakes. When the reader isn’t confident, you see a note saying some words may be wrong, so check names and numbers.

Tips for better results

• Take the photo straight on, in good even light, with the text filling most of the picture.
• Crop away everything but the text with Crop image. Pictures, logos and edges of the table can turn into stray characters.
• For a tilted photo of a page, Document scanner straightens it from its four corners. Save as JPG per page, then read that picture here.
• Got a PDF instead? Scanned PDF to text (OCR) reads every page of a scanned PDF.