How to fix sideways PDF scan pages

Scanned pages lying sideways: turning them the right way up

The sheet feeder flipped some of the pages, the phone caught the document in a different orientation, the book was scanned as spreads. The upshot is the same: part of the file lies wrong and has to be read with your head tilted.

The operation is simple, but it has three quirks that account for almost every unexpected result.

Rotation is a page property, not rotated content

Inside a PDF the page does not turn over. The content stays lying the way it was, and a note is stored alongside it: "display this page turned by such and such an angle".

The main practical consequence follows from that. Different programs treat that note differently. Your viewer honours it and shows the page correctly, while the reader built into somebody else's document system may ignore it and show the original orientation.

So check the finished file somewhere other than where you worked. Drag it into a new browser tab: a browser's reader behaves closest to what runs inside intake forms and portals.

Rotations accumulate

The second consequence of the same design: a rotation does not set an orientation, it is added to the current one.

A page already turned 90 degrees becomes 180 after a second 90-degree turn. That is exactly the case where somebody runs the job again because "it did not work last time" and ends up with an upside-down document.

It is easy to verify: measured on a file where the first pass set 90 degrees on page one, and an identical second pass made it 180.

If you have already turned a page and want it back, turn it 180, or 90 the other way. There is no separate reset button.

Where wrongly oriented pages come from

There are only a few causes, and the pattern of the mess usually tells you which one fired.

The scanner's sheet feeder. Some sheets go into the tray the other way round and land in the file at their own angle. The tell: individual pages are flipped at random while the rest are fine.

Double-sided scanning without duplex. The stack is run face up, flipped, and run again for the backs. The backs then sit 180 degrees off relative to the fronts. The tell: exactly every second page is upside down.

A phone shot. If the picture carries no orientation tag in its metadata, the page lands on its side. The tell: the photographs are the ones on their side while the scanned sheets stand upright.

Landscape inserts. A wide table or a drawing inside a portrait document is often saved rotated so it fits. That is not a mistake, and usually should be left alone: it was made that way to be read by turning the screen.

That last case is worth separating out before you start. Turning a landscape insert "like the rest" leaves you with a table that has to be read sideways.

How to turn the pages

1. Open rotate-pdf and upload the document. 2. Wait for the preview grid and find the pages lying wrong. 3. Tick only those. 4. Choose the angle and the direction. 5. Run the job and check the result in a browser.

Steps two and three matter more than they look, as described below.

With no pages selected the whole document turns

If no page is ticked, the rotation applies to all of them.

That is convenient in exactly one situation: the scanner laid the entire document on its side. Then nothing needs ticking and one operation fixes everything.

In every other situation it is a trap. In a forty-sheet scan, three or four are usually wrong, and running with nothing selected lays the other thirty-six on their sides. The mistake is not noticed immediately, because the first page happens to look right.

Tick by the picture of the page on the grid, not by numbers you remember. That is what the preview grid is for.

The direction outranks the sign of the angle

The parameters hold both an angle and a direction, and they can contradict each other. That is resolved unambiguously: the direction wins.

The angle you type is taken as an absolute value, and the chosen direction gives it its sign. Verified: an angle of minus 90 together with the clockwise direction turns the page clockwise, not counterclockwise.

The practical conclusion: do not try to express direction with the sign of the angle. Set the direction from the list and type a positive number.

Angles that are not multiples of 90

A PDF can only turn a page by right angles: 90, 180, 270. An arbitrary number in the angle field is rounded to the nearest multiple of 90.

There is a non-obvious edge to that. The value 45 sits exactly halfway between 0 and 90, and it rounds towards zero, that is, produces no rotation at all. If you typed 45 and nothing changed, that is why, not a malfunction.

For a tilt of a few degrees, rotation is the wrong tool entirely. A skewed scan is cured by straightening rather than turning: rebuild it through scan-to-pdf, which has real deskewing based on the page content.

Double-sided scans: why turning half the pages does not help

A separate scenario in which rotation looks like the answer and is not.

Scanning a stack without duplex produces a file holding all the fronts first and all the backs after, often in reverse order and upside down. Turning the backs 180 degrees is not enough: they are still in the wrong places.

Two operations are needed, in sequence. First the rotation of the pages that sit upside down. Then a reorder, so each back follows its own front.

That order specifically. Reordering is easier once every page reads normally: the grid shows content rather than a set of inverted rectangles.

What carries over into the result

Document properties and bookmarks survive: rotation changes only the orientation of pages and does not touch the composition of the document.

The page count does not change. If it did, you ran the wrong tool.

The text layer stays in place and keeps working. Selecting text with the mouse on a rotated page does behave oddly in some programs, because the selection box is built along the content's original orientation. Search is unaffected.

The order relative to other operations

Rotation belongs before recognition and before compression.

Before recognition because ocr-pdf reads the page image, and rotated text is recognised noticeably worse than upright text.

Before compression because compression rebuilds the document, and an extra pass over it after rotation achieves nothing.

Cropping with crop-pdf is also more logical after rotation: the margins there are set as you see the page on screen, and on a rotated page they are remapped automatically, but they are easier to think about once the page stands upright.

And if the pages are not turned but out of order, the tool is a different one: order belongs to organize-pdf.