Fixing sideways and upside-down scans (and why they’re rare now)

Old scanning routine: scan, sigh, rotate, save again. ToPDF runs orientation detection before first display, so pages nearly always appear upright — even ID cards photographed sideways in one hand. Here's the system and its edges.

How the page knows which way is up

After cropping, a small on-device model classifies the page among 0°, 90°, 180°, 270° — learning from text layout the way you do at a glance — and the page is turned before it ever renders. No sideways flash, no manual step. (More on the model.)

The pages that can still fool it

• Nearly text-free pages: photos, pure diagrams, blank forms — little signal to read.
• Decorative/rotated typography: certificates set in ornate script at angles.
• Mixed-direction layouts: tables with vertical labels dominating.

For these, tap Rotate — and note your correction is sticky: once you set a page's rotation manually, automation stops second-guessing that page, including after re-crops.

Landscape content is not “wrong”

A landscape table is correct in landscape; orientation detection aims for readable-as-printed, not portrait-at-all-costs. In a mixed PDF each page keeps its own correct orientation — see scanning landscape documents.

Frequently asked questions

Why was this historically such a mess?

Cameras record an orientation flag that different apps honour differently, and scanners guessed from the flag rather than the content. Reading the actual page content sidesteps the whole mess.

Does rotating afterwards hurt quality?

No — 90° rotations are lossless re-arrangements of pixels.