Words in a photo should be searchable.
- What you need
- A photo of a menu, a page or someone’s post is only a picture until you can find it by what it says.
- What I decided
- I read the text on the device, with Apple’s Vision framework, when you scan with the camera or pick a photo. Nothing is uploaded to do it. The text goes into the note, so Search finds the note by what the picture says, and any web address in it becomes a link.
- What it costs
- It is only as good as the photo, because there is no cloud pass to rescue a blurry one. If nothing is readable, the note is still saved, with a dated title and no text, instead of an error.
How it works
- It runs on camera scans and photo-library picks, on expense screenshots and PDFs, and on scanned PDFs in document import. A PDF with its own text is read directly. OCR only steps in when the PDF has almost none, for up to 30 pages.
- Scans ask Vision for 19 languages with automatic detection, Arabic and Chinese among them. Expense imports ask for four: English, Chinese (simplified and traditional) and Arabic, a narrower list than the scanner’s.
- The result is tidied before it is saved. Lines are ordered top to bottom, paragraph breaks come from the gaps between lines, and hyphenated English words are rejoined. Arabic is left in the order Vision returns it.
- On a social screenshot, interface text such as like counts, Follow, Reply and timestamps is stripped out and the platform is recognised. The title is the first line of what was read, cut at 60 characters, and a confidence level (high, medium or low) is stored with each scan.
- Expenses found by OCR wait in a review inbox, with possible duplicates flagged, before they become entries.





























