What is OCR and how has it changed?

OCR — optical character recognition — is the technology that reads the shapes of letters and turns them back into machine-readable text. It's been around for decades (Tesseract, the open-source engine most free tools use, dates to the 1980s) but for most of that history you had to install it, train it, and tolerate middling results on anything but clean printed text.

Three things changed in the last five years:

• WebAssembly matured. Tesseract was compiled to WASM as Tesseract.js, meaning you can run a real OCR engine inside any browser tab at near-native speed.
• LSTM models replaced older pattern matching, boosting accuracy on real-world screenshots substantially (Tesseract v4+).
• Browsers got fast enough that recognizing a full A4 page takes 2–5 seconds on a modern laptop — faster than uploading it to a server in many cases.

The net result is that you can do quality OCR without installing anything, without uploading anything, and without paying anything.

Option 1: Browser-based OCR (recommended)

Tools like extractmytext.app, onlinetext.org, and similar run the whole OCR pipeline in your browser. Your image is decoded, preprocessed, and recognized locally. Nothing is uploaded.

Best for: screenshots, photos of documents, product labels, memes, menus, printed pages, anything containing sensitive information. Results are instant after the first language pack downloads.

Limits: a one-time 5MB–15MB language model download per language. Handwriting, extreme fonts, and very stylized text are weaker than specialized commercial services like Google Cloud Vision.

Option 2: Cloud OCR services

Google Cloud Vision, AWS Textract, Microsoft Read API, and ABBYY Cloud offer the best absolute accuracy available — particularly for handwriting, tables, and multi-column layouts. They also have free tiers (e.g., 1,000 pages/month on Google Vision) that are plenty for personal use.

Trade-off:you're uploading your image to a third party. For sensitive data, that's a non-starter. You also need an account, an API key, and usually a bit of code to call the service — or rely on a wrapper site that might log your image.

Option 3: Desktop apps

Adobe Acrobat Pro, ABBYY FineReader, and macOS Preview all include OCR. They are excellent if you already own them and need to OCR many PDFs. For a quick one-off screenshot, installing a new app is overkill.

Step-by-step: using extractmytext.app

Step 1 — Open the tool

Go to extractmytext.app. No signup, no paywall, no popup.

Step 2 — Add your image

You have three options, pick whichever is fastest:

• Drag and drop the image file onto the dashed drop zone.
• Click the drop zone to browse. PNG, JPG, WebP, BMP, and GIF all work.
• Paste from clipboard with Ctrl+V (Windows/Linux) or Cmd+V (Mac). This is ideal when you just took a screenshot — the screenshot tool on every major OS saves to clipboard by default.

Step 3 — Pick a language

English is selected by default. If your image is in another language, switch to it in the dropdown. The first time you use a language, extractmytext.app downloads a ~5MB training data file. Your browser caches it, so next time it's instant.

Mixed-language images (say, an English article with a few French phrases) usually work fine under English mode. If the secondary language has accents the English model handles them.

Step 4 — Wait a few seconds

The progress bar shows you exactly what's happening. You'll see stages like "Loading language data", "Initializing OCR engine", and "Recognizing text". A typical screenshot takes 2–5 seconds total on a modern laptop; a full scanned page takes 10–20.

Step 5 — Review and clean up

The extracted text appears in an editable textarea. You'll also see the confidence score. Anything above 85% is usually publication-ready; 70–85% needs a quick proofread; under 70% means the input image needs improvement.

Step 6 — Export

Three options: Copy to clipboard, Download as .txt (plain text), or Download as .md (includes the source filename, language, and confidence score as a header — handy for notes apps like Obsidian or Bear).

Accuracy tips that actually move the needle

What about handwriting?

Handwriting is hard. Tesseract-based tools (including extractmytext.app) work well on neat printed handwriting — block letters, all-caps notes, whiteboard sketches — but cursive and messy handwriting are unreliable. For critical handwriting OCR, Google Cloud Vision's handwriting mode or Microsoft OneNote's built-in OCR are the best free options.

What about PDFs?

extractmytext.app is for images, not PDFs directly. If you have a PDF, export the page you need as an image (every PDF viewer can do this), or screenshot it. Most PDFs from the last decade already have a text layer you can copy-paste — only scanned PDFs need OCR.

What about HEIC from iPhone?

HEIC files aren't decodable in most browsers. Convert HEIC to JPG or PNG first using convertmyheic.app (also free and client-side), then drop the result into extractmytext.app.

Why client-side OCR wins for most people

If you're OCRing anything with sensitive content — a medical form, a passport scan, a private message, a signed contract — cloud OCR creates a real risk. The file sits on someone else's server, possibly logged, possibly retained.

Client-side OCR eliminates that risk entirely. Your image is never transmitted. The whole workflow happens in the same tab you opened, and when you close the tab the image is gone. For most common OCR tasks — extracting text from a screenshot, a photo, a scan — the quality is more than good enough.

FAQ