📄 OCR - Extract Text from Images
Upload an image and extract text using Tesseract.js OCR. Supports 10 languages. Recognition runs in your browser after the OCR engine downloads from a CDN.
Last updated: November 8, 2026 · By the BoltQuickTools team
Pull text out of a screenshot, scan or photo
Error messages in screenshots, receipts, photographed pages and scanned forms hold text you can read but cannot select. This page reads the image for you so you can copy the text instead of retyping it.
Running a recognition
- Choose the text's language in the Language menu. Each run uses one language model.
- Add an image: drop it on the upload area, click the area to browse, or press Ctrl+V anywhere on the page to paste from the clipboard.
- Click Extract Text. The progress bar names each stage while the engine loads and reads.
- The result appears under Extracted Text with a Confidence badge and word and character counts. Use Copy to Clipboard to take it, or Clear to start again.
A test image and what came back
A 640 by 160 pixel PNG with black 40px Arial on white, reading "Invoice 2041" on one line and "Total due: $18.50" on the next, was run with Language set to English. The tool returned:
Invoice 2041
Total due: $18.50
The page showed Confidence: 95%, Words: 5, Characters: 30. The character count includes the line break. Clean digital text like this is the easy case. Expect lower confidence from photographs.
What travels over the network
The image itself stays in your browser. The page reads it into memory and hands it to a Tesseract.js worker, and the page's code never uploads it. The engine is not bundled with the site, though. Tesseract.js is loaded from the jsDelivr CDN, and it then downloads its worker, its WebAssembly core and the model for your chosen language. You need a connection unless your browser has those files cached, and the first run in a language takes longer. A fresh worker is created for each click and shut down when it finishes.
Where it struggles
- One language at a time. Words in any other language come back as the closest guess in the selected model. Chinese is Simplified only.
- Layout is not kept. You get lines of text. Tables and columns collapse, and no fonts or formatting carry over.
- Confidence is an average, not a guarantee. The badge is coloured high at 80 and above, medium from 50 to 79, and low below that. A single wrong digit can sit inside a high score, so check amounts, codes and dates by eye.
- Handwriting and skewed photos read poorly. Very small text helps from being enlarged first, for example with the image upscaler.
- Only the first file counts when you drop several, and a dropped file that is not an image is ignored without a message. There is no size limit in the code, so a huge image is limited only by your device.
PDFs, bad output and batches
Can I run it on a PDF?
No, the upload only takes image files. If the PDF has selectable text, copy it directly. For a scanned PDF, take a screenshot of the page and paste that here. PDF Tools can pull the pages you care about into a smaller file first.
Why is the output empty or nonsense?
The tool shows exactly what the engine found. Nonsense usually means the Language menu does not match the image, or the text is too small or too low in contrast. An empty result simply reports 0 words.
Can I process a batch of images?
Not on this page. Each new image replaces the current one, so a set of images means one extract per image. Paste the results into the word counter if you need totals across them.