Needs argument validation and some other changes. Needs testing with mixed-content PDFs. Only really works for pure invisible text at the moment.