Image to Text (OCR)
Reads printed text from photos and screenshots. Handwriting is a different problem — the page explains why it rarely works.
Drop an image here or click to choose
Photo, scan or screenshot · nothing is uploaded
Why handwriting usually fails
OCR recognises letter shapes it has been trained on, and printed type is consistent in a way handwriting is not. Tesseract, the open-source engine this uses, is trained overwhelmingly on printed text and typically manages 95%+ accuracy on a clean scan. On handwriting it usually produces nonsense, because every writer forms letters differently and cursive joins them into shapes with no clear boundaries. Handwriting recognition is a genuinely harder problem needing different models — so if that is what you need, this is not the tool.
Getting a usable result
| Image | Typical accuracy |
|---|---|
| Screenshot of on-screen text | Near perfect |
| Flatbed scan at 300 dpi | Very good |
| Straight, well-lit phone photo of a page | Good |
| Angled or shadowed photo | Patchy |
| Text over a photograph or pattern | Poor |
| Handwriting | Usually unusable |
Four things help most: square on rather than angled, even light with no shadow across the page, in focus, and large enough — characters need roughly 30 pixels of height, so fill the frame with the text rather than the whole desk.
What is downloaded, and what is not
OCR needs a recognition engine and trained language data. This page fetches Tesseract.js and the language file you select from public CDNs the first time you use them — several megabytes, cached afterwards.
Your image is not part of that. Recognition runs in WebAssembly on your own device. The picture never leaves your machine, which matters given that the things people most often run OCR on are documents, receipts, letters and ID.
Most free OCR websites upload your image to a server instead.
How to use the Image to Text (OCR)
- Drop in an image, or just paste a screenshot with Ctrl+V.
- Pick the language of the text — this matters more than you would expect.
- Press Read the text. The engine downloads once per language.
- Proofread and edit the result, then copy it.
Frequently asked questions
Is my image uploaded?
No. The OCR engine and language data are downloaded to your browser, and recognition runs locally in WebAssembly. Your image is never transmitted.
Does it read handwriting?
Rarely well. The engine is trained on printed text; handwriting needs different models entirely. Neat block capitals sometimes work; cursive essentially never does.
Why is the first run slow?
The engine and the language data download on first use, a few megabytes. After that they are cached and subsequent runs are much faster.
Can it read a PDF?
Not directly. Use the PDF to image tool first, then run the resulting page image through here.
Why is the text full of mistakes?
Usually a low-resolution, angled or shadowed image. Retake it square on with even light, filling the frame with the text. The confidence figure tells you how much to trust the output.
Does it keep the layout?
Roughly. Line breaks are preserved but columns, tables and complex layouts come out as running text, which usually needs rearranging by hand.
Related tools
Last reviewed September 4, 2026. Results are estimates for general information only.