Turn a screenshot or document photo into copyable text
OCR stands for optical character recognition. It estimates which characters appear in an image and returns text that you can copy into an email, document or other application. This is useful for screenshots, printed notes, receipts and scanned paperwork that cannot be selected like normal text.
Choose a language that matches the document. An English/Urdu or English/Hindi combination is useful when both scripts appear on a page. Recognition can vary significantly with font, script, print quality and layout; the language choice is not a promise of perfect transcription.
PDF text extraction and OCR are different
Many PDFs already contain selectable text. With Prefer existing selectable text enabled, the tool extracts that layer when it is present and runs OCR on pages where it is missing. This can be much faster than recognising every page as an image.
If the PDF has an incorrect or incomplete hidden text layer, turn the preference off to recognise the displayed page instead. Plain-text output does not preserve page layout, table structure, fonts or columns as an editable Word document. Text order may need correction on complex layouts.
A PDF range can contain up to 20 pages per run. For large files, process a smaller range and download those results before continuing. Password-protected PDFs must be opened or exported as an accessible copy first.
Get more accurate recognition
- Use a sharp, well-lit source with sufficient text size.
- Straighten photographed paperwork with the Document Scanner first.
- Choose the correct language and orientation.
- Check names, dates, amounts and reference numbers manually.
- Treat handwriting, unusual fonts, tables and damaged scans as difficult inputs.
The displayed OCR confidence comes from the engine and is only an estimate. It is not a measured accuracy percentage. The result editor lets you correct the text before downloading it. Changing the text does not recalculate the original engine confidence.
Images are reduced to a longest edge of up to 3,000 pixels for processing. PDF pages are rendered at a suitable scale within the same limit. Recognition cannot recover characters missing from the source image.
Private recognition and model downloads
The OCR runtime and selected language models are served by UploadWell and run in your browser. The first use requires a download; later visits can benefit from browser and language-model caching. Large scans can take longer on older phones.
Your files and extracted text are not sent to an OCR service. The working results remain in the current page, so download or copy them before closing it. Processing can be stopped; completed results stay available.
If your next task is completing a form rather than transcribing it, try Fill & Sign PDF. For image-based copies with better page geometry, use the Document Scanner.