Image to Text Converter

Extract text from screenshots, scans and photos in your browser. Free OCR for English, Chinese and Japanese, many images at once, nothing uploaded.

Your files are processed in your browser and never uploaded.

Line breaks

Drop images here

Choose files

HEIC, JPG, PNG, WebP, AVIF, GIF, BMP · up to 50 megapixels each · you can also paste

Works with screenshots, scans and photos of printed text in English, Chinese and Japanese. Handwriting is not supported. The AI models (about 16 MB) download once, then stay in your browser's cache.

    How to use

    1. Drop one or more images, or click "Choose files". Screenshots, scanned pages and phone photos all work.
    2. The text is found and read line by line. Choose whether to keep the line breaks as in the image or join lines into paragraphs.
    3. Correct anything in the text box if needed, then copy the text or download it as a .txt file.

    Features

    Accurate on real pages
    A modern text detector and recognition model read printed text in any layout, from a chat screenshot to a scanned contract, and keep the order of lines and paragraphs.
    English, Chinese and Japanese
    Latin script with accents, simplified and traditional Chinese, Japanese kana and kanji, digits and symbols, mixed freely on the same page.
    Paragraphs or exact lines
    Keep every line break as it appears, which suits lists, addresses and code, or join wrapped lines back into paragraphs for pasting into a document or an email.
    Private and unlimited
    Recognition runs on your device. Images are never uploaded, there is no account and no daily quota, and you can convert a whole folder of screenshots in one go.

    How the text is recognised

    Recognition happens in two stages, using models from the open-source PaddleOCR project. The first, a text detector called PP-OCRv5, looks at the whole image and marks every region that contains a line of text. It works on images in any layout: a single caption, a dense page, or text scattered over a photo. If the picture is slightly tilted, the image is straightened by the average slant of the lines it found.

    Each line is then cut out, scaled to a fixed height and read by PP-OCRv6, a recognition model with a vocabulary of about 18,000 characters. The model reads the whole line at once rather than letter by letter, so it uses context: a character that is ambiguous on its own, such as 0 or O, is usually resolved by the characters around it. Lines with very low confidence that consist of only one or two characters are dropped, which removes most of the specks and edges that a photo can contain.

    Putting the text in order

    The recognised lines are arranged in reading order. Lines that sit at the same height belong to one row and are joined from left to right; rows are listed from top to bottom. A gap between two rows that is clearly larger than the normal line spacing starts a new paragraph. If you choose to join lines into paragraphs, wrapped lines are joined with a space, Chinese and Japanese lines without one, and a word split by a hyphen at the end of a line is put back together.

    Tips for better results

    Use the sharpest image you have: a screenshot is better than a photo of a screen, and a photo taken straight on in good light is better than one at an angle. If the text is small, zoom in before you take the screenshot. Crop away anything that is not text with the crop tool, and turn multi-column pages into one image per column. To keep the original images together with the text, combine them into a document with JPG to PDF.

    Frequently asked questions

    What is OCR?

    OCR, optical character recognition, turns a picture of text into text you can select, edit and search. It is what you need when a document only exists as a screenshot, a scan or a photo, for example a receipt, a slide, a page of a book or a message someone sent you as an image.

    How good is the recognition?

    Clean screenshots and scans are read almost perfectly. Phone photos work well when the text is sharp and evenly lit. Typical mistakes are look-alike characters, such as a lowercase l and a capital I, or a stray symbol picked up from a smudge. Everything appears in an editable box, so you can fix the odd character before you copy the result.

    Can it read handwriting?

    No. The models are trained on printed and on-screen text. Very neat block capitals sometimes work, but cursive handwriting and notes generally do not.

    Does it keep the layout?

    It keeps the reading order. Lines go from top to bottom, words on a line from left to right, and a blank line between paragraphs. Formatting such as fonts, bold or colour is not kept. For tables, use [Image to Excel](/en/image-to-excel/), which rebuilds rows and columns. Pages with two or more text columns side by side are read across the whole width, so it is best to crop each column separately.

    Is there a limit on images or file size?

    No daily limit. Images up to 50 megapixels are accepted, in JPG, PNG, WebP, AVIF, HEIC, GIF or BMP format. Large batches are read one or two images at a time so your device stays responsive.

    Are my images uploaded to a server?

    No. The page downloads the AI models once (about 16 MB) and every image is read in your browser. That makes it suitable for private material such as ID cards, bank statements or work documents.

    Related tools