To look for a specific photo or PDF, you simply have to type in a keyword or phrase like you would on a search engine. Dropbox will then show you the files that contain those words or phrases. The company told VentureBeat that this is "the most computationally intensive project its machine learning team has ever undertaken." They were particularly challenged by PDFs, since multi-page documents require a lot more processing power than an image file. In order to make indexing them feasible, they designed the system to stop extracting and indexing text after 10 pages.
Automatic image text recognition works for English-language JPEG, static GIF, PNG, TIFF and PDF files on Dropbox, even those uploaded before the service rolled out the feature. However, it's availability is fairly limited. Dropbox Business Advanced and Enterprise users might be able to access it soon, depending on when their account administrators switch it on, while Dropbox Professional subscribers will get the feature in the coming months. Ordinary users will have to keep on looking for documents the old-fashioned way.