How to Extract Text
from an Image (OCR)
Tesseract.js running in your browser. 100+ languages, no upload, no signup.
Open Image OCRWhen you actually need OCR
You took a photo of a recipe in a cookbook. You have a screenshot of someone's slide deck. You scanned a contract that came in as a PDF of images. You photographed a handwritten whiteboard. You want to copy a quote off a meme. Every one of these starts as pixels and ends with you wanting plain text — that's OCR.
Browser-based OCR with Tesseract.js is fast enough for these everyday cases without the privacy hit of uploading your screenshots to Google Vision or AWS Textract. The model runs locally; nothing ever leaves your device.
Step-by-step: image → text
Open the Image OCR tool
Visit the free Image OCR tool. It runs Tesseract.js — the WebAssembly port of Google's open-source Tesseract OCR engine — entirely in your browser.
Drop in your image
Drag a JPG, PNG, GIF, or BMP. Screenshots, photographed pages, scanned documents, receipt photos — all work. Higher resolution gives better OCR; aim for at least 200 pixels per character of text.
Pick the language
Default is English. Switch to French, German, Spanish, Chinese, Japanese, Korean, Russian, Arabic, or any of 100+ supported languages. Each language model downloads on first use (~5-10 MB) and is cached for next time.
Wait for OCR to run
First run takes 5-30 seconds depending on image size and your CPU. The progress bar shows phases: loading model, recognising layout, recognising characters. Subsequent runs are faster because the model is cached.
Copy or download the text
Click Copy to grab the extracted text, or Download as .txt. Line breaks roughly preserve the original layout; you may need to clean up artifacts on noisy or low-resolution sources.
Pro tips for better OCR accuracy
Resolution matters more than file size
200+ pixels per character of text is the sweet spot. A blurry 4 MP photo of a page reads worse than a sharp 1 MP screenshot. If your phone photos look soft, take them again with the macro mode or steady the device.
Crop before OCR for cleaner output
Tesseract works best on tight crops with minimal background. Crop to just the text area before uploading, especially for receipts or page corners. The Image Splitter or any crop tool helps with this preprocessing.
Black-on-white beats every other contrast
Tesseract was trained primarily on book pages. White text on dark backgrounds, neon signs, or low-contrast designs all reduce accuracy significantly. If you control the source, prefer well-lit black-on-white.
Handwriting is mostly hopeless
Tesseract's printed-text models don't handle handwriting reliably. For handwriting, use a service with a dedicated handwriting model (Google Vision, AWS Textract). The free Tesseract approach works only on cleanly typeset text.
Confidential documents stay confidential
Photographed contracts, scanned ID documents, screenshots of internal Slack — all stay on your device. Tesseract.js runs entirely as WebAssembly inside your browser tab.
Frequently asked questions
What is OCR?
Optical Character Recognition — turning an image of text into machine-readable text. The image goes in (a photo, a scan, a screenshot); plain text comes out. OCR engines have existed since the 1970s; modern ones use neural networks for layout analysis and character recognition.
How accurate is browser OCR?
On clean, well-lit, printed text: 95-99% character accuracy. On noisy phone photos or low-contrast designs: 70-90%. On handwriting or stylised fonts: usually much worse. Accuracy scales sharply with input quality — clean source = clean output.
What languages are supported?
100+ languages with native scripts including Latin, Cyrillic, Greek, Arabic, Hebrew, Devanagari, Chinese (simplified and traditional), Japanese (with hiragana/katakana/kanji), Korean (hangul), and Thai. Each downloads its model on first use.
Why is the first run so slow?
Tesseract has to download the language model (5-10 MB), initialise the WebAssembly module, and load the model into memory. After that it's cached — second-run latency is just the actual recognition (a few seconds for a typical screenshot).
Can I OCR a multi-page PDF?
Not directly — convert each PDF page to an image first using PDF to Images, then OCR each one. For pure text-extraction needs (PDF that already has selectable text), use PDF OCR which handles the conversion automatically.
Is the image uploaded to a server?
No. Tesseract runs in your browser as WebAssembly. Image data goes from your file picker into Tesseract's input pipeline directly — never through a network. Even the language model downloads come from a CDN once and then sit in your browser cache.