Skip to content

Documents and text in photos

Frameleaf reads the text in your photos: receipts, signs, letters, screenshots, anything with words in it. You can search by that text, copy it, and correct it where it was misread.

Documents, under Explore in the sidebar, lists your photos that show text. A photo is listed because of the text read from it. The search box finds photos by that text, or by your own corrections to it.

Text in photos is also part of the main search.

Open a photo and its Info panel ends with Text in this photo. It shows:

  • the text as it reads now
  • Show text regions, which draws where each line is on the photo
  • Select text and Copy text

If the photo is yours, you can also review its text:

Action What it does
Correct Fix a line that was misread. Your correction is kept beside the recognised text, never over it, so the original reading stays on record
Dismiss Hide a line that’s just noise. Restore brings it back
Show where Points at the line on the photo. Hovering over or focusing a line does the same
Read text again Runs text recognition on this photo again

Read text again uses the text recognition destination your administrator chose. If text recognition is switched off, or no destination is chosen, it tells you so instead of sending the photo anywhere.

People you share the photo with see the text with your corrections applied and without the lines you dismissed. Only you see your review itself.

  • Read again. Your corrections follow the line that now covers the same place in the photo. If the new reading differs from the text you corrected, the line says so. A correction whose text has gone entirely is kept as your own text.
  • Cropped. Text that a crop removes is no longer read, listed or searchable, and your corrections to it are hidden while the crop hides it. Search on a cropped photo only matches its visible text and values you typed yourself. Remove the crop and it all comes back.
  • Locked. A Locked photo’s text is part of the photo. It’s listed and readable only while your session is unlocked, and locking again clears it from the list and from an open Info panel.

If your administrator turns on Suggest receipt and document fields (in the OCR part of the machine learning settings, off by default), the panel suggests details read from the photo’s text:

  • a date
  • a total
  • a reference
  • an email address
  • a phone number

A suggestion is only what the text says. It shows where it was read and how confident the recognition was, and it’s never treated as a verified record: a clearly read “12.50” can still be the wrong total, and the order of day and month in a date isn’t guessed.

For each suggestion you can:

  • Confirm it, or pick one of the Other readings
  • Correct it to your own value
  • Dismiss it
  • Reset your decision so the suggestion shows again

Every change you make is checked against the version of the text you were looking at. If the photo was read again, or you changed the same line on another device in the meantime, your change is refused, the panel reloads, and nothing is overwritten. Make the change again on the fresh text.

Your corrections travel with your photos in a preservation package, and are restored where your library has none of its own.