OCR that actually works: extract text from images and scans
Image to text turns pictures of writing into text you can select, search and edit. The quality of the photo matters more than most people expect.
Prepare the image first
- Straighten the page. Even a small tilt can make the text harder to read.
- Crop closely. Remove the desk, fingers and page edges so the tool sees only the text.
- Raise contrast. Dark text on a light, even background wins; use Image filters to push brightness and contrast, or convert to grayscale.
- Use a clear image. Small or blurry letters are much harder to read correctly.
- Avoid shadows and glare. Flat, diffuse light beats a bright lamp at an angle.
- Don't over-shrink. Resizing a scan down before OCR throws away the detail the engine needs.
What to expect from each option
| Option | Strength | Trade-off |
|---|---|---|
| Toolsii image to text | Free, private, reliable on clean printed text | Struggles with handwriting and complex tables; first run downloads a language model |
| Online professional services | Best on handwriting, receipts, low-light photos | Costs money per page and uploads your document |
| Text already inside a PDF | Perfect accuracy — it is real text, not a picture | Only works if the PDF was generated digitally, not scanned |
Try Extract PDF text first on any PDF. If it returns nothing, the pages are images — export them with PDF to images and OCR each page.
Fixing the output
Extracted text can contain unwanted hyphens, broken lines, and mistakes such as l/1, O/0 and rn/m. Paste the result into AI OCR cleanup to rejoin words and restore paragraphs, then proofread numbers by hand — those are the errors a language model cannot detect.
FAQ
Is the OCR free and private?
Yes. The image is read inside your browser — it is never uploaded, and there is no page limit or signup.
Which languages work?
Ten languages ship with the tool, including English, Arabic, French, German, Spanish and Portuguese. Choosing the right one matters more than any other setting.
Can it read handwriting?
Not reliably. Tesseract is trained on printed type; neat block capitals sometimes work, cursive rarely does.
Why is the first run slow?
The language model downloads once and is then cached, so subsequent pages are much faster.
Tools: Image to text (OCR) · AI OCR cleanup · Extract PDF text