MultiKit

Image and PDF to Text (OCR): Extract Text in Your Browser

  • Free
  • No sign-up
  • Runs in your browser
  • Version 1.0.2
  • Last updated 2026-10-25
Report a problem

Drop a picture or a PDF and get its text back. A PDF that already holds text is copied exactly; a scan or a photo is read with OCR, all inside your browser.

    The first reading in a language loads its data (1 to 3 MB) from this site, and the browser keeps it for next time.

    Everything you enter is processed in your browser and is never sent to a server.

    Browsers:Chrome119+Edge119+Firefox121+Safari17.4+Opera105+Samsung Internet25.0+Browsers

    How to use

    1. Choose or drop pictures (JPG, PNG, WebP, HEIC, SVG) or PDF files, up to 20. You can add more files later and remove ones you do not want.
    2. Pick the language of the text. If it also contains English, leave Also read English on. Then press Read text. The first reading in a language loads its data from this site, which takes a few seconds.
    3. The text appears below as each picture or page is finished, and you can stop at any point. Edit it in the box if you like, then copy it or download it as a .txt file. Mark where each page starts adds a line with the file name and page number.
    4. Use Send to under the text to pass it to another tool, for example Text Compare, Case Converter or Character Counter.

    Examples

    A screenshot of an error message, language English
    → The message as plain text, ready to paste into a search or a bug report
    Screenshots are sharp and high-contrast, which is the best case for OCR.
    A 3-page PDF, 2 pages typed in a word processor and 1 page scanned
    → Pages 1 and 2 are taken from the PDF text; page 3 is read by OCR
    The summary line tells you how many parts were read by eye and how many came from PDF text.
    A photo of a Korean receipt, language Korean with English
    → The shop name, items and prices as lines of text
    Photos taken straight on, in good light, read far better than slanted or blurred ones.

    How it works

    The reading is done by Tesseract, an open-source OCR engine that began at Hewlett-Packard and is developed by Google and many contributors, running in the browser as WebAssembly through the Tesseract.js library. It does not look for exact matches of letter shapes; a neural network of the LSTM kind reads each line of the picture and predicts the characters. The language data used here are the “best” models, stored as integers to keep them small.

    Before reading, a picture is drawn onto a white canvas at a good size: small pictures are enlarged up to twice, very large ones are reduced so that their longest side is about 4,000 pixels, and the turn that phone cameras record is applied. For PDFs, each page’s own text is read first with pdf.js. A page with no text, or only a few stray characters, is drawn at about 300 dpi and read by OCR.

    OCR output is a best guess. The tool is tested on text drawn into a picture, on a PDF that holds text, and on a PDF that is only a picture, but real photos vary, so check numbers, names and anything important against the original.

    Get a more accurate OCR result

    • Best input: a straight, well-lit JPG, PNG or PDF with sharp, high-contrast printed text. Choose the document’s language before reading.
    • Output: copy or download plain text. A text PDF is read directly; image-only pages use OCR.
    • Privacy: files, rendered PDF pages and recognized text stay on this device while the language model is served from MultiKit.
    • Limits: handwriting, decorative fonts, curved pages, tables and multi-column reading order can be wrong. Crop or straighten a photo first, and verify names, dates and amounts against the source.

    Sources

    Frequently asked questions

    How do I get the text out of an image?
    Open the image here, choose its language and press Read text. The reader looks at the picture the way a person would and writes down the letters it recognizes. Then copy the text or download it as a .txt file.
    How do I extract text from a PDF?
    Add the PDF. If its pages hold real text, as PDFs made from documents or web pages do, the text is taken from the file itself, which is exact and fast. If the pages are pictures, as in scans, each page is drawn at about 300 dpi and read by OCR.
    How accurate is it?
    For clear, printed text of a normal size it is usually very good, with a few mistakes to proofread. Accuracy drops with handwriting, small or blurry text, strong slants, busy backgrounds, and photos taken at an angle. Choosing the right language matters a lot.
    Which languages can it read?
    English, Korean, Japanese, Simplified Chinese, Spanish, French, German, Portuguese and Russian. Mixed text works when you keep Also read English on.
    Is my file uploaded anywhere?
    No. The reader and its language data are served from this site and run in your browser, so your pictures and PDFs never leave your device.
    Why does the first reading take longer?
    The language data, 1 to 3 MB, is loaded the first time you read in that language, and the reading engine starts up. The browser keeps the data afterwards, so the next readings start sooner.
    How can I get better results?
    Use a sharp, straight, well-lit picture where the letters are at least about 20 pixels tall. Crop away borders and backgrounds, which the Crop Image tool can do, and send the result here with Send to. Do not use a picture that is much larger than needed.
    Can it read a password-protected PDF?
    Not directly. Remove the password first, for example by opening the PDF in a viewer and saving a copy without it.
    Version history
    • 1.0.2 — Adds a practical decision guide covering best-fit uses, inputs and outputs, privacy, quality trade-offs and known limits.
    • 1.0.1 — Makes the search title and description state the supported task, key formats and privacy benefit more clearly.
    • 1.0.0 — First release.

    Last updated: