The inclusion of text in the Seadragon demo was demoware that hasn't made it into a product unfortunately. For the foreseeable future Photosynth will only deal with images. So -- if you really want a page from a document to synth in, you need to "print" it to a bitmap. Not great I know.
Also -- you should know that our current matching algorithms often produce false positives on text. Two different pages of text that happen to have some lettershapes in the same place show up as matches -- causing problems. In the future we plan to integrate text recognition into the synther which would remove these false positives.