Extract text from a scanned PDF
Scans become text you can copy and search, without the document leaving your device.
Drop files here
or click to choose from your deviceAccepted: PDF
One file per runUp to 20 MB per file5 runs per dayRemove limits with Pro✓ Works offline
Your files stay on your device
This tool runs entirely in your browser. Nothing is uploaded, nothing is stored, and you can even turn off your internet connection before you start.
Example
Example: a 14-page lease where 12 pages are copied and 2 scanned pages are read
lease-signed.pdf came back from the landlord as one file: pages 1–12 are the lease exported from the property manager’s software, with real text, and pages 13–14 are the pet addendum, printed, signed and scanned. You want all of it as text for your notes. Language of scanned pages: English, European languages, Japanese, Chinese.
- Leave Pages empty. Pages 1–12 carry their own text, so it is copied exactly and at once; only pages 13 and 14 go through text recognition.
- Only those 2 scanned pages count toward the 20 a run can read, so the whole lease is read in one run.
- The first time, the recognition models (about 11 MB) download; each scanned page is then rendered at 150 dpi and read in a few seconds.
- The text is marked --- 1 --- to --- 14 ---, so a clause you find in it points straight to its page in the PDF.
- To read just the addendum again, type 13-14: the file is then lease-signed-p13-14.txt, marked --- 13 --- and --- 14 ---.
| Pages | What is read | Text file |
|---|---|---|
| empty | 1–12 copied, 13–14 recognised | lease-signed.txt, --- 1 --- to --- 14 --- |
| 13-14 | 13 and 14, recognised | lease-signed-p13-14.txt |
| 3 | page 3 only, copied | lease-signed-p3.txt, no page marker |
One .txt to copy or download; the PDF itself is not changed. Check names, dates and dollar amounts from the scanned pages against the scan: recognised text can miss a character, copied text cannot.
How it works
- Drop a PDF: a scan, a photographed document saved as PDF, or any PDF you want the text of.
- Leave Pages empty to read the whole file, or enter pages such as 1-3, 5. Up to 20 scanned pages are read per run.
- Press Extract text. The first time, the text recognition models and their runtime (about 11 MB) are downloaded.
- Copy the text from the box or download it as a .txt file. Pages are marked --- 1 ---, --- 2 --- and so on.
Frequently asked questions
- Does it make the PDF searchable?
- This page gives you the text. To keep the PDF and make it searchable, use Make PDF searchable, which adds the same recognised text as an invisible layer over each scanned page.
- What happens to pages that already have text?
- Their text is taken straight from the PDF, which is instant and exact. Only pages without usable text, such as scans, go through text recognition.
- Why only 20 scanned pages at a time?
- Reading a page takes several seconds and a lot of memory in the browser. For a longer scan, enter the pages in batches, such as 1-20 and then 21-40.
- Which documents work?
- Printed Japanese, English, German, French, Spanish, Italian, Portuguese and Turkish. Japanese can be horizontal or vertical; vertical pages are recognised automatically and are a little less accurate, with 、 and 。 sometimes left out. In Turkish, ı and İ are sometimes read as i or I. Handwriting is not read reliably, and horizontal pages set in several columns can come out with the columns mixed.
- Do pages that already have text count toward the 20?
- No. Only pages without usable text are counted and recognised; pages with their own text are copied, however many there are. A page with fewer than 30 characters of its own text, such as a scan carrying only a page number, counts as scanned.
- What is the text file called?
- The PDF’s name with .txt. When you read only some pages, they are added to the name: lease-signed-p13-14.txt for pages 13–14, report-p1-3_7.txt for pages 1–3 and 7.
- Is my PDF uploaded?
- No. Pages are rendered and read in your browser. The free plan handles 1 PDF per run, up to 20 MB, and 5 runs per day.