Recognize text in a scanned PDF or replace text that copies as gibberish. Select, review, and copy words without uploading your document.
Your files stay on your device. No account required.
or drop a PDF here
Select to open here. Examples use synthetic content.
Choose the current page or draw a box around just the text you need.
OCR runs in your browser. Models download only when you start recognition.
Select recognized text, correct spelling or numbers, and keep or join line breaks.
Open a PDF, recognize the current page or draw around one passage, then select the words you need. Review the result and fix mistakes before copying. Your document stays on your device.
Open your document once. Switch between Text, Table, and Images without reopening it.
Choose a local document or try the scanned invoice example. You do not need an account.
Choose Recognize page, or Select area and draw around a passage before choosing Recognize area. The first recognition downloads about 49 MB of model and runtime files.
Highlight the recognized words on the PDF. Edit the preview, choose how to handle line breaks, and copy the result.
The built-in example is a fictional invoice rendered into a page image. It has no selectable PDF text layer. Choose it above and run OCR to read its labels, item descriptions, and amounts.
A rasterized invoice — visible text, without selectable characters.
Select recognized words on the page, review the preview, then copy. Recognition may need corrections.
Designed for clear printed English and Chinese text. Quality varies with language, font, contrast, orientation and scan resolution. Review names, amounts, identifiers and punctuation. Handwriting, complex columns, sideways text and degraded scans can read poorly. A selected passage is often easier to check than a whole page.
Recognition processes one page or area at a time, with a limit of 250 detected text regions and a bounded image size. Large pages may be reduced in resolution. You can switch between original and recognized text. This tool does not create a searchable PDF, unlock encrypted documents, or export an editable PDF.
OCR reads table text but does not reconstruct scanned table rows and columns in this version. Table mode continues to use the PDF’s original text layer.
No. Recognition runs in a browser worker using WebAssembly. Your device downloads the OCR files, but the app does not upload your PDF, page images, or recognized text.
Yes, if the PDF can be rendered clearly. OCR reads the visible page pixels rather than relying on its font-to-character mapping. It does not repair the original PDF and may introduce recognition mistakes.
Your browser first downloads and initializes the OCR runtime and models. Later requests may reuse normally cached website assets. Each recognition also takes time depending on your device and the amount of text.
Yes. Select an area before recognition, then select individual words from the recognized text. Character highlight positions are approximate.
No. This version provides selectable recognized text and a copyable, editable text preview. It leaves the original PDF unchanged.