Split PDF
Drop one PDF and get its pages back as separate files, in this tab. Every output is a new document built from copied pages: the page content, its own size and its rotation come through, while bookmarks, form fields and any link that pointed at another page stay with the file you started from. The panel reports how many files the split produced, what they cost in bytes against the source, and the page range inside each one. Nothing is uploaded.
Drop a PDF here
or choose a file · one PDF, up to 100 MB / 2,000 pages · or paste with Ctrl + V
One PDF in · up to 200 files out · every page keeps its size and rotation · bookmarks, forms and cross-page links do not travel · no upload · no account
Split PDF: What One Page Costs When It Becomes Its Own File
A 15-page paper weighing 2,215,244 bytes was cut three ways in this tab. Split into two files, its pages weighed 1,950,716 bytes together, 11.94% below the original. Split into fifteen, the same pages weighed 2,857,147 bytes, 28.98% above it. The page count never changed, and neither did the pages: page 1 rendered to 1,090,584 pixels with 0 of them different, and its 2,802-character text layer came back with the same hash.
That gap is the part a split screen usually leaves out. Splitting does not divide one file into pieces. Each output is a new document, and pages are copied into it, so every new file pays for its own copy of whatever those pages touch.
Every output starts as an empty document
The library used here opens the source, reads the page objects you asked for, and adds them to a document created from nothing. The new document begins with an empty catalog, so the structures that live at file level are not carried over:
- Bookmarks. The paper carries 22 outline entries. Every one of the 15 single-page files has 0.
- Title and author. A test file saved with the title "Quarterly report" and the author "A. Person" produced single-page outputs where both fields were empty. The producer and creator lines were replaced with the library's own name.
- Form definitions. A text field with widgets on two pages was gone from both outputs, even though the widget rectangle stayed on the page (more on that below).
- Page labels. Numbering such as i, ii, iii is a document-level table, and no output writes one.
What remains is the page tree. That is why a split is a copy job with a smaller scope than "divide the file": the pages you keep are rebuilt, and everything that pointed at them from outside is left in the original.
The bill scales with the number of files, not the page count
The same 15 pages, cut six ways:
| Output files | Total bytes | Against the 2,215,244-byte source |
|---|---|---|
| 2 | 1,950,716 | −11.94% |
| 3 | 2,025,329 | −8.57% |
| 4 | 2,119,778 | −4.31% |
| 5 | 2,184,710 | −1.38% |
| 8 | 2,395,095 | +8.12% |
| 15 | 2,857,147 | +28.98% |
The two-file split and the fifteen-file split contain the same pages, in the same order, with the same content. The difference between them is 906,431 bytes, or 46.47% of the two-file total. Totals move by a few bytes between runs; the percentages hold.
Two mechanisms drive the curve. A page does not own its font or its images; it points at them in shared resource dictionaries. When a page is copied into a new document, the objects it points at are copied too, because the new file has no way to reach back into the old one. A page that shares a font with fourteen others therefore carries a copy of that font in every output that contains it. On top of that, each output writes its own header, object table and trailer, which is a fixed cost per file.
The direction is not always upward. Two, three, four and five files all came out smaller than the source, because the source's own file furniture is dropped and the copies are written compactly. Eight files crossed over into growth, and fifteen added 29%. If you need the smallest possible result, fewer files win; if you need one page per file, the pages themselves were never the expensive part.
One page weighs whatever its resources weigh
The 15 single-page files ran from 40,803 bytes (page 12) to 538,126 bytes (page 14), a 13.2× spread. Page 11 and page 12 sit close together in the document and both come out near 40 KB; pages 13 to 15 each carry figure-heavy content and land between 332 KB and 538 KB.
So "split into single pages" does not produce equal-size files, and a mail server with a 100 KB attachment cap will accept some pages and reject others from the same document. The panel lists the byte size of every output before you download, which is the only reliable way to know.
What survives the copy
Four things were checked page by page, not assumed:
- Pixels. Page 1 was rendered before and after the split at 1.5×, giving a 918 × 1,188 canvas. Both renders held 1,090,584 pixels and 0 differed; the largest channel difference was 0.
- Text. The same page's text layer held 2,802 characters in both files, with the same SHA-256 prefix. Page 9 held 2,922 characters in both. Nothing is rasterized, so the words stay selectable and searchable.
- Size. All 15 pages are 612 × 792 pt in the source, and every output page reports the same.
- Rotation. A 400 × 300 pt page rotated 90 degrees came out of a split with /Rotate 90 intact, so it still displays as 300 × 400.
- Link rectangles. The source holds 113 link annotations, 30 of them on page 2 and 18 on page 10. The 15 single-page files held 113 between them, none dropped.
A link that pointed elsewhere has nowhere to go
A three-page test file was built with a link on page 1 whose destination was page 3. Page 1 was split into a file of its own. The annotation survived, rectangle and all, but the destination now resolves to page index 0, which in a one-page file is the page you are already on. The words that said "see page 3" are still there; the place they pointed at is not.
The paper's 99 internal destinations behave the same way after a page-by-page split: those annotations stay on their pages, but every one of them points at a page that lives in a different file. External URLs are unaffected; 14 of the annotations are ordinary web links, and a URL does not need the document around it.
This is not something the panel can repair. A destination is a reference to a page object, and rewriting it would mean guessing which new file now holds the target. If the links matter, split by ranges that keep linked pages together.
A form field turns into an empty box
A 300 × 300 pt page was built with a text field named customer_name and the value "Ada", with two widgets, one on each of two pages. After a two-way split, each single-page output reported 0 form fields. The widget annotation was still on the page: a rectangle where the field used to be, with no name, no value and nothing to type into.
The reason is the same split between page level and document level. The widget is drawn by the page; the field that gives it a name, a type and a value lives in the document's form dictionary, and a page copy never visits that dictionary. A reader that opens the output sees the box. A reader that expects a fillable form sees nothing.
What this page will not do
- Merge. Multiple inputs are refused; one PDF goes in.
- Compress. A split changes the container, not the page images. The output can be larger than the source, as the table above shows.
- Reorder or delete pages. Page order inside a file follows the source.
- Extract text or OCR. Words are copied, not recognized.
- Remove a password. An encrypted PDF is refused before a page is read; the panel names it and stops.
- Repair a damaged file. A PDF that cannot be parsed is reported, not fixed.
- Keep bookmarks, forms, metadata or cross-page links. Those stay with the source, as described above.
- Take an unlimited file. The ceiling is 100 MB and 2,000 pages, with 200 output files per run, because every byte stays in this tab's memory.
How to use this page
- Drop one PDF onto the panel, or paste it with Ctrl + V.
- Choose how to divide it: every page, every N pages, or a page list.
- The split runs on its own after each change; the result block reports the file count and total size.
- Download a single PDF directly, or a ZIP when the split produced more than one file.
What each control does
Split by sets the unit. Every page makes one file per page. Every N pages makes equal chunks; the slider and the number box stay in step, and the hint shows how many files that will produce. Page list takes a string such as 1-3, 5, 8-10 and writes one file per comma-separated group, in the order typed, which also covers plain page extraction.
The result block lists each output with its page count and byte size, plus the total against the source and the time the split took.
Download hands back a PDF when there is one file and a ZIP when there are several. The ZIP is written in the tab with no library, and its entries are the same bytes the panel measured.
Why this page doesn't upload your file
The PDF library is served from this site and runs in the tab. Reading the source, copying pages and writing the outputs all happen in the browser's memory, so no request carries a document out. With the network cut after the library loaded, the same 15-page paper split into the same 15 files, with the same page counts and sizes and a total within 3 bytes of the online run.
Questions that come up
Why does splitting into more files make the total bigger? Each output is a complete PDF with its own copy of the fonts and images its pages reference, plus its own object table. Two files share those resources across many pages; fifteen files copy them fifteen times. The measured difference on one 15-page paper was 906,431 bytes.
Can I keep the bookmarks in the split files? Not from this page. An outline is a document-level tree, and a new document starts without one, so 22 entries became 0 in every output. A desktop editor that edits the outline after splitting can rebuild it; a browser copy cannot carry it.
What happens to a link that points to another page? The link rectangle stays, but its destination points at a page that is no longer in the file. In a one-page output the destination resolves to page 1, so clicking it keeps you where you are. Split ranges that keep linked pages together if the links have to work.
Why is one single-page file larger than another? A page pays for the resources it draws with. Pages 11 and 12 of the test paper came out near 40 KB; pages 13 to 15, which carry figures, came out between 332 KB and 538 KB. The panel reports each file's size before you download.
Does each page keep its size and rotation? Yes. All 15 pages of the test paper kept 612 × 792 pt, and a page rotated 90 degrees kept its rotation in the output. Nothing is rescaled to a common paper size.
What do I do with a password-protected PDF? This page cannot open it, because the pages cannot be read without the key. Open it once in the application that owns the password, export an unprotected copy, and split that copy.