pdfpix

Turn a PDF into HTML

or drop one here
Layout

Keeps page sizes, text positions, fonts, drawings and images. Text remains selectable.

Choose a PDF first.

Convert a PDF into one HTML file without uploading it. Preserved pages keeps the page layout, selectable text, fonts, paths and images. Choose Flowing article for headings, lists, tables and linked footnotes that reflow. Both files open without scripts or fetched assets. The download requires a text check; that check does not prove pixel equality.

How to turn a PDF into HTML

  1. Choose a PDF

    Press Choose a PDF, or drop one on the page. The file is opened in this browser tab and the tool reports its page count before you start.

  2. Choose the layout

    Preserved pages is the default. It keeps the page sizes, positioned text, fonts, paths and images. Choose Flowing article to rebuild paragraphs, headings, lists, tables and linked footnotes instead.

  3. Compare the preview

    The original first page and the HTML appear side by side. Preserved pages keep headers and page numbers; Flowing article removes repeated page furniture. Check the appearance before using the file.

  4. Check and download

    The finished HTML is handed back to the browser’s own parser and its text counted against the text the PDF held. A failed check does not produce a download.

Two layouts for different uses

Preserved pages keeps the PDF's coordinate system. Text stays selectable SVG text with embedded fonts, formulas keep their glyph positions, and paths and supported images keep their place on the page. Page sizes and rotations survive, including mixed-size documents. The pages scale down on a narrow screen; paragraphs do not reflow.

Selectable text is not the same as semantic HTML. Preserved pages does not reconstruct heading levels, table cells or an accessible reading order. Choose Flowing article when you need document structure and paragraphs that fit the screen.

Flowing article takes the structure from the PDF itself when the PDF has it. A tagged PDF carries a tree of headings, lists, tables and figures, and that tree is followed where it reaches the text. Most PDFs are not tagged, and for those the order is rebuilt: columns are found from the gaps between text, lines are joined into blocks, and the blocks are read down each column rather than straight across the page. The receipt says which route ran, so you know whether the structure came from the file or from the measurement.

Headings are decided against a measured body size rather than a fixed threshold. The most common text size in the document is the body, the sizes above it become the heading levels in order, and a document that uses one size throughout gets no invented headings. Lists keep their nesting and are written as ul and ol with their items intact rather than as paragraphs that start with a bullet character.

Tables are found twice over. A table drawn with rules is read from the rules, which is the easy case. A table held together by nothing but the spacing between its columns is read from the column positions, which is the case most converters flatten into lines of text. Both are written as a real table with its caption, its header row marked th scope="col" and its first column marked th scope="row", so the result is navigable rather than just square. A table that runs over a page break is joined back into one.

Footnotes are linked rather than dumped. A marker in the body becomes an anchor marked role="doc-noteref", the note becomes an aside marked role="doc-footnote", and the two share an id, so the jump works in a reader and in a screen reader. Notes collected at the back of a document are written as one endnotes section instead. Markers that point at nothing and notes that nothing points at are both counted on the receipt rather than hidden.

Running furniture comes out. Text that repeats in the same position across pages is the running head, the running foot and the page number, and it belongs to the paper rather than to the document. It is found early and removed late, after the tables are settled, because a repeated table header and a "continued" caption repeat in the same way and must not be mistaken for it. Every removal is counted and the counts appear on the receipt.

Flowing article does not carry pictures. Its output is markup and nothing else: no images, no scripts, no addresses to fetch, one small stylesheet written inline. That is deliberate, and it is why the preview can be shown in a frame with every permission switched off. When you need the images from the same PDF, Extract images pulls them out at their stored resolution.

A scan is refused rather than half converted. With no selectable text there is nothing to read, so the tool says so and offers no download instead of producing an empty document.

Both layouts parse the finished HTML back before offering a download. Preserved pages also compares the identities and counts of non-space characters on each page against a separate PDF text read. Flowing article requires at least 98% retention overall and 90% per page after deliberate removals, with the emitted text in order.

These checks do not compare pixels. Font substitution in a PDF that lacks embedded fonts, antialiasing, thin strokes and image resampling can differ between viewers. Preserved pages refuses drawing features it cannot handle, including Type 3 fonts, soft masks and text clipping. It never silently switches to Flowing article.

How to convert PDF to HTML covers when this tool is the right route and what to do with a scanned PDF, and How to convert HTML to PDF covers the way back.

Questions

Does PDF to HTML upload my file?
No. The PDF is read, converted and checked in this browser tab. Neither the PDF nor the HTML is uploaded.
What does the HTML look like?
Preserved pages writes one SVG per page, with selectable text, embedded fonts, vector paths and embedded images. Pages scale to fit the window without changing their layout. Flowing article writes headings, paragraphs, lists, tables and linked footnotes that reflow with the window. Both produce one file with no script and nothing to fetch when it opens.
How are footnotes handled?
Preserved pages keeps footnotes where they were drawn. Flowing article links each detected marker to its note and groups detected endnotes at the back. Its receipt reports unmatched markers.
Are tables kept as tables?
Preserved pages keeps the text, rules and cell positions, but does not infer HTML table semantics. Flowing article detects tables from their rules or aligned columns and writes table elements with marked headers.
What happens to images?
Preserved pages embeds supported images in the HTML alongside the selectable text and vector drawing. Flowing article does not carry images. Extract images can save the embedded pictures separately.
What about a scanned PDF?
It is refused rather than half converted. A scan holds a picture of a page and no selectable text, so there is nothing to turn into markup. The tool says so at the point it finds out and offers no download. Run OCR on it first, then convert.
Does it remove the running head and page numbers?
Not in Preserved pages. Flowing article detects repeated headers, footers and page numbers, removes them after detecting tables, and counts those removals on the receipt.
What does the check on the finished HTML prove?
It checks text, not pixel equality. Preserved pages compares non-space character identities and counts against a separate PDF text read on every page, then parses the HTML back. Flowing article requires at least 98% text retention overall and 90% per page after deliberate removals, and checks the emitted text order. Unsupported drawing features or failed checks prevent a download.

The other pdfpix tools

All 37 of them run in a browser tab like this one. Each reads its own output back and checks it before offering the download, and none of them needs an account.

Where to go next