A scanned contract with one wrong number you need to fix before signing. The text isn't selectable — it's a picture of text.
OCR converts the scan into recognisable text. Then editing becomes possible.
Start with a clean scan
OCR works best on straight, well-lit pages. Before running OCR, deskew the scan if it's crooked.
Low-quality scans (low contrast, faded ink, smeared text) give poor OCR. If accuracy matters, re-scan at higher quality.
Run OCR
Open the PDF in Flint's editor and run OCR. The tool analyses every page and adds an invisible text layer.
The visible scan looks identical. Behind it, the text is now recognised and editable.
Edit the recognised text
Click any text in the document. If OCR worked, the text becomes editable. Type to replace, delete, or insert new text.
Key caveat: OCR isn't perfect. Check the recognised text before editing — sometimes "0" comes through as "O", "1" as "l". Fix any recognition errors before relying on the text.
Save with the new text
After editing, save the file. The exported PDF has both the original scan and the edited text layer. Most readers show the edited version.
For full clean-up, re-export with just the text layer — but be aware that loses the original scan visual.
FAQ
How accurate is OCR?
90-95% for clean printed text. 80-90% for low-quality scans. Always proofread for accuracy on key fields (numbers, names, dates).
Can OCR handle multiple languages?
Yes, if you specify the language. Mixed-language documents work but with some accuracy loss.
Will my edits be visible immediately?
Yes, in most readers. Some older readers show only the original scan and miss the text layer edits.
What if OCR misreads important fields?
Correct the recognised text in the editor before treating it as final. OCR errors in legal or financial fields are critical to catch.
Can I edit handwritten scans?
OCR for handwriting is much less accurate (40-70%). For handwritten text, treat OCR as a search aid only, not for editing.
A scan becomes a real document with OCR plus editing. Open the file in Flint's editor, recognise the text, then fix what needs fixing.