PDF Tools

A PDF is not a document the way a word-processor file is. It is a fixed-layout description of marks on a page — text runs carrying their own coordinates and font references, images, vector paths — assembled from numbered objects and indexed by a cross-reference table near the end of the file. That design is why it renders identically everywhere, and why editing one feels awkward: there is no paragraph to reflow, only glyphs that have already been placed.

Almost every job in this category is therefore structural rather than editorial. You are moving whole pages between files, changing how a page is oriented, or translating the page description into something else — pixels or plain text. Knowing which of those you need explains most of the confusion about why one tool worked and another seemed to do nothing.

Which tool for which job

Two libraries doing two different jobs

Structural work — merging, splitting, extracting, rotating, building from images — is handled by pdf-lib, which reads and writes the PDF object graph directly. Work that needs a genuine interpreter — rendering a page, pulling out the text, reading the metadata — uses pdf.js, the same engine that displays PDFs inside Firefox and Chrome, and it loads only when a tool asks for it.

The distinction shows up in the results. A merge never rasterises anything, so it is fast and lossless: page content streams are copied intact and the text in the output is still selectable, searchable text. A conversion to JPG has to interpret every glyph and paint it, which is a much heavier operation and produces pixels you can no longer select.

Merging keeps order and gives up some metadata

Merge PDF concatenates whole documents in the order you list them, so arranging the files is the ordering step — there is no re-sorting afterwards. Page content is copied rather than re-rendered, which means vector artwork, embedded fonts and links inside a source document survive.

What does not survive is per-document metadata. The output carries one title, author and producer record, because a merged file cannot honestly claim three different authors. Expect form fields and digital signatures to be dropped or invalidated too: a signature covers a specific byte range, so any rewrite breaks it. Sign after merging, never before.

Splitting and extracting are the same operation with different intent

Split PDF breaks one file into several; Extract PDF Pages keeps only the pages you name. Both accept the same range syntax — 1-3, 5, 8-last, where first, start, last, end and * are all understood — and page numbers outside the document are ignored rather than treated as a failure, so one typo does not cost you the whole run.

Splitting is also the usual way to get under an email or portal attachment limit without wrecking legibility. One file per page gives the largest total, because every output has to repeat the font and resource objects its pages depend on; chunks of five or ten pages normally compress far better.

Rotating sets a flag, it does not redraw

Rotate PDF writes the page's /Rotate entry in 90° steps and leaves the content stream untouched, which is why it is instant and lossless. A sideways scan is fixed properly rather than papered over.

One consequence is worth knowing. A downstream system that ignores /Rotate and reads raw coordinates still sees the unrotated page. If a portal rejects your scan as landscape when your own viewer shows it upright, render it with PDF to JPG and rebuild with JPG to PDF — that bakes the rotation into the pixels permanently.

PDF to image: resolution, memory and the white background

PDF to JPG renders each page onto a canvas at a scale you choose. PDF's default user space is 72 units per inch, so 1× is roughly 72 dpi, 1.5× about 108, 2× about 144 and 3× about 216 — the last being real print territory. The longest edge is capped at 4,096 pixels regardless, because a 3× render of a large-format sheet would ask the browser for a canvas it cannot allocate.

Pages render one after another rather than in parallel. That is slower and deliberate: parallel renders exhaust memory on a phone, and each canvas is released as soon as its image is encoded, since Safari holds GPU memory for a canvas until it is resized to zero.

PDF pages are transparent by definition, so every page is flattened onto white first. Without that step a JPEG — which has no alpha channel at all — would come out black. Pick PNG instead of JPG when the page is mostly text and you want the edges to stay crisp; the image tools cover what to do with the result.

Building a PDF from photographs

JPG to PDF and HEIC to PDF place one image per page, scaled to fit inside a margin, on A4, US Letter, A3, A5, US Legal, or a page sized exactly to the image. Orientation can be forced or left on auto, which chooses portrait or landscape per image from its aspect ratio — usually what you want for a mixed batch of phone photos.

Only JPEG and PNG can be embedded in a PDF directly. Anything else — WebP, GIF, SVG, HEIC, or a file whose extension does not match its contents — is decoded and re-encoded first, so the output is a genuine PDF rather than a renamed picture. Captions and centring are optional, and the document title comes from the first filename, which you may want to correct afterwards with PDF Metadata.

Text extraction stops where OCR begins

PDF to Text reads the document's actual text layer and reconstructs line breaks from the vertical position of each text run. For anything produced by a word processor, an email client, a browser's print-to-PDF or a design tool, the result is accurate, fast, and real editable text.

For a scanned PDF it returns nothing, and that is not a malfunction. A scan is a photograph of a page: there are pixels where the letters are, and no text objects anywhere in the file. Recovering characters from pixels requires optical character recognition, which this tool does not perform. You can tell which kind of file you have in two seconds — try selecting a word in any PDF viewer. If you cannot select it, there is no text layer to extract.

The tool reports which pages came back empty, and that list is the quickest way to find a scan hiding in the middle of an otherwise digital document.

What compression can and cannot do

Compress PDF rewrites the file using cross-reference object streams and clears the title, subject, keywords, producer and creator records. On a document written by software that emits older, verbose structures, that reclaims real space: object streams typically remove a few percent on their own, and stale metadata plus the accumulated revisions of incremental saves can account for more.

It does not recompress or downsample the images inside, and it does not discard embedded fonts. A file that is large because it contains thirty high-resolution scans will barely change, and implying otherwise would waste your time. Check what you are holding first with PDF Metadata. If the images are the problem, render the pages at 1.5× with PDF to JPG, compress those with the image compressor, and rebuild — a lossy round trip, but one that will actually reach an attachment limit. The tool always shows input and output size side by side, so you can see straight away whether it helped.

Metadata is the part people forget to check

PDF Metadata reads the document information dictionary: title, author, subject, keywords, the application that created the file, the producer library that wrote it, creation and modification dates, page size and count, and whether the document is encrypted.

Before a file leaves your organisation, that list is worth reading. Author fields often hold a personal name or an internal username. Titles carry draft filenames nobody meant to publish. Creation and modification dates establish a timeline. And the producer field names the exact software and version you used, which is a small but real disclosure about your internal tooling.

Password-protected files

A PDF can be protected in two quite different ways. An owner password restricts permissions — printing, copying, editing — while leaving the content readable by anyone who opens the file. A user password encrypts the content and is required to open it at all.

The first kind opens normally here, so merging, splitting and rotating a permissions-restricted document works. The second cannot be opened without its password, and these tools will say so rather than produce something corrupt. There is deliberately no password-removal tool in this category.