Drop a PDF and get the text out. Pages that already contain text are read instantly and exactly; only the scanned, image-only pages go through recognition. The file never leaves your machine.
A PDF can hold text in two completely different ways. If it was exported from a word processor or an accounting system, the characters are really in the file — those pages are read out directly, character for character, with no recognition and no error.
If the page was scanned or photographed, it is just a picture of writing. There are no characters to read, which is why you cannot select the text. For those pages the scan is lifted straight out of the file at its original resolution and passed through optical character recognition — slower than the first case and, unlike it, able to misread. The list above tells you which method each page used.
Your document is opened by your own browser and stays there. No request carries it, nothing is stored, and no account is needed — which matters when the PDF is an invoice, a contract or a bank statement.
To be exact about the one network request that does happen: the first time you use the page, your browser downloads the recognition engine from a public code repository, the same way it downloads any script. After that it is usually cached and the page works offline. The site also counts anonymous page views. Neither of those touches your file.
Running a whole invoice through recognition to get one IBAN or one total is a lot of work for one number — and you still have to find it in the output. Snip2Field sits on your Windows desktop: press a hotkey, drag a box around just that field, and it reads it, checks an IBAN against its checksum, and copies it. Fully offline, like this page.
See Snip2FieldNo. The file is opened and processed by your own browser. It is never sent to a server, and nothing about its contents is stored. The page downloads its recognition engine from a public code CDN the first time you use it, and that is the only network request involved.
Because the page is an image rather than text. Scanned, photographed or image-exported PDFs contain pixels that look like letters but carry no characters to select. Recognition turns those pixels back into text, which is what this tool does for the pages that need it.
Pages that already contain real text come out exactly, with no recognition involved. For scanned pages, accuracy depends on the scan: clean 300 dpi text is usually very good, while low-resolution photos, unusual fonts and handwriting are much less reliable. Always check numbers against the original.
Neither. Because the work happens on your machine there is nothing to meter. Very long scanned documents simply take longer, since each image page is recognised one at a time.
After the first run, usually yes — the browser caches the recognition engine. The first visit needs a connection to fetch it.
Recognition can misread a digit in any tool, which matters far more for a bank number than for prose. The IBAN checker will tell you whether an IBAN you extracted is internally valid, and Snip2Field runs that check automatically when it copies one.