Extract Text from Images and PDFs (OCR) - Free, No Upload
Turn a photo, scan, or PDF into copyable, searchable text. Recognition runs entirely in your browser tab using the open-source Tesseract engine - the image never leaves your device.
Maintained by Hirantha Sandaruwan Lakshan. Updated 31 July 2026.
How it works
The first time you run this tool, your browser downloads a small text-recognition engine and an English language model (about 6 MB total, cached afterward so it's instant on later visits).
Each page is analysed for lines, words, and characters, and the recognised text is assembled in reading order.
For a PDF, every page is rendered as an image first, then read in sequence - the result keeps a page separator so you can tell where each page's text starts.
Worked examples
A photo of a whiteboard or printed page
Drop a JPG or PNG straight from your camera roll and get back plain text you can paste into notes, an email, or a document.
A scanned PDF with no selectable text
Many scanned documents are just images wrapped in a PDF - this pulls the actual words out so you can search, copy, or edit them.
Quoting text from a screenshot
Instead of retyping a paragraph from a screenshot, extract it directly and paste it wherever you need it.
Frequently asked questions
How accurate is it?
Very good on clean, well-lit printed text and typical scans. Handwriting, low resolution, heavy skew, or unusual fonts reduce accuracy - straightening and cropping the source image first helps a lot.
Does it only work in English?
Right now, yes - only the English text model is bundled. Mixed-language pages will still attempt recognition, but non-English words are more likely to come out wrong.
Why does the first run take longer than the second?
The recognition engine and language model download once and are cached by your browser. Every run after that skips the download entirely.
Is my image or PDF uploaded anywhere?
No. Rendering, recognition, and text extraction all happen locally in your browser tab.
Can it keep the original layout, like columns or tables?
No. The output is plain text in reading order. Columns and tables lose their structure, which is the normal trade-off for text extraction without a layout engine.
Does it work on handwriting?
Poorly. The engine is trained on printed type. Neat block capitals sometimes come through; ordinary cursive generally doesn't.
What image resolution works best?
Enough that the text is comfortably readable when you look at it - roughly 300 DPI for a scan. Too low and characters blur together; far higher mostly just slows things down.
Can I get a searchable PDF instead of plain text?
Not currently. You get the recognised text, which you can paste wherever you need it.
Sources & references
Updated Jul 31, 2026