Notices by fnetX (fnetx@social.anoxinon.de)

@clacke Interesting thought, but I think in this case it's rather an issue with text extraction: In the first two images, text is well separated with plenty of space around it.
The lower two images don't have that much whitespace.

I'd assume it didn't properly separate the images and mixed text line-by-line between both images.

You can see a correlation between the "William" ... name and the "hope" (which is near each other visually), and the "little Billy" ... and the "messed up now".

@clacke You could try manipulating the image and inserting some more whitespace between the two lower images, or aligning them below each other.

My guess is that it would properly extract the text in this case.