Out of a document

Make a PDF searchable

The same pages, with the words underneath.

On this device

Choose a scanned PDF, or the pictures of its pages

One PDF, or any number of pictures in page order. The reader is downloaded the first time you use it.

    Opened on this device · never transmitted

    What this tool does, exactly

    A scan is a picture of a page, and a viewer cannot search a picture. This tool reads every page and writes the words back over it in a font that draws nothing, so the document looks exactly as it did and answers a search. A pile of photographs of pages becomes one searchable PDF the same way.

    What it does
    Draws each page at 200 dpi, finds and reads every line of text on it (PaddleOCR's PP-OCRv6 with ONNX Runtime, in this tab), and writes each line invisibly over its place on the original page — rendering mode 3, a one-glyph font, the construction Tesseract's PDF output has used since 2013. The original page content is sealed and left as it was; only a layer is added.
    What it checks
    Before the file is offered it is opened again with the same PDF library this site reads with, and every page's layer is read back and compared, line by line, with what was written. A page that does not read back is named in the result; the file is still yours, because the scan in it is intact.
    When it refuses
    When the PDF already has a text layer of more than about twenty characters a page. A second, guessed layer over a real one would give every search two answers, and the real one was there already.
    Pictures
    JPEG and PNG go in as they are; WebP, AVIF, GIF and BMP are converted once. Each picture becomes an A4-proportioned page, in the order you chose them. Photographs of pages read best when they are flat and well lit — the Scan to PDF tool straightens and cleans them first.
    Which languages
    English and forty-six other Latin-script languages, Chinese and Japanese by default; Russian, Ukrainian and Belarusian with a second recogniser you choose in the options, an extra eight megabytes.
    What it costs to load
    About seventeen megabytes compressed the first time — the page renderer, the reader and its two models — fetched only when you open this tool and cached afterwards.

    About this tool

    Will the PDF look any different?

    For a scanned PDF, no: its pages are kept as they are inside the file, and the layer is written in a font with one blank glyph and in a rendering mode that paints nothing. A viewer shows the scan and, when you search or select, finds the words at the places they were read from. Pictures are different: each is fitted onto an A4-proportioned page, and WebP, AVIF, GIF and BMP are re-encoded once as JPEG.

    Will it work in any PDF viewer?

    It is the construction Tesseract has written into PDFs since 2013 — a Type 0 font with an Identity encoding and a ToUnicode map, nothing the PDF standard has not had since 2001. Here it is read back with pdf.js before the file is offered, and pdftotext reads it with the layout intact; other viewers were not measured, and a viewer that cannot search this file cannot search Tesseract's either.

    Can I get the text on its own as well?

    Use OCR a PDF, which gives you the searchable PDF and the text and Markdown together from the same reading. This tool exists for the person who only wants the searchable file and does not want to wade through outputs to find it.

    How accurate is the hidden text?

    As accurate as the reader: on a clean 200 dpi scan of printed type, measured at 0.2% of characters wrong in English, under 1% in Portuguese and Russian. A search for a word the reader misread will not find it, so the result panel lists the lines the reader was least sure of.

    Is the document uploaded?

    No. The reader and its models come to this browser; the document never leaves it. The whole job — drawing, reading, writing the layer and reading it back — runs in this tab.

    More out of a document tools · All 51 · How they work