Point it at a photo of a page and get editable text back. OCR is genuinely useful, but only for certain kinds of images. Here is the honest boundary.
Open the tool →| Source | Expected accuracy |
|---|---|
| Flatbed scan, 300 DPI, straight | Excellent — near-perfect for printed text |
| Phone photo, flat page, good light | Good — minor errors, check numbers |
| Phone photo, angled or shadowed | Poor — expect missing characters |
| Handwriting | Unreliable — usually not worth it |
| Screenshot of text | Excellent — clean pixels |
Before running OCR, do three things: make the page flat, make the lighting even, and increase the contrast. Most OCR failures are image-quality failures, not engine failures.
OCR confuses characters that look alike: 0/O, 1/l/I, 5/S, rn/m. Numbers in invoices, account numbers and dates are where errors cost money. Always verify digit-heavy sections by eye before trusting the output.
Poorly. OCR engines are trained on printed type. Neat block capitals sometimes work; cursive almost never does.
JPG, PNG and WebP. Very large images may need resizing first so the browser can process them.
No. OCR extracts a stream of text. Columns, tables and text boxes come out as plain paragraphs — you will need to reformat in Word.
No. Recognition runs in your browser. Check the Network tab in DevTools while it works.
Compress PDF · Merge PDF · Split PDF · PDF to Images · Images to PDF · PDF to Word · Word to PDF · Remove Watermark · Resize Images · Image Converter · Temp Mail
Make a scanned PDF searchable
Turn a scanned image into editable Word text
What is a PDF file, really?