OCR for mobile upload iOS only works when upside down

[julisu]

Describe the bug

I am using the latest version. When taking photos to upload them, ocr will not work. As soon as I turn the document upside down the ocr works perfectly.
I am using the upload feom taking a photo on my iPhone on iOS 26.6.1

What happened?

Is there a solution so I don’t have to turn every document on its head?

System information

System:
OS: Linux 7.0 Debian GNU/Linux 13 (trixie) 13 (trixie)
CPU: (2) x64 unknown
Memory: 1.78 GB / 2.00 GB
Container: Yes
Shell: 5.2.37 - /bin/bash

Where did you encounter the bug?

A self hosted instance

[Xadoy]

I was trying to investigate the same problem with an agent. And found the following:

Can confirm - same root cause here. iPhone photos store rotation in the EXIF orientation tag instead of rotating the actual pixels. Browsers honor that tag, so the preview looks upright, but Papra pipes the raw buffer to Tesseract (packages/lecture/src/tesseract/tesseract.usecases.ts) which ignores EXIF and has no OSD enabled (default PSM 3), so it reads the physically inverted pixels and outputs non-readable wierd text (since it's upside down).

Likely fix: run image buffers through sharp(buffer).rotate() (no args = applies EXIF orientation) before OCR. sharp is already a dependency of @papra/lecture. Happy to test a patch.