[hdjjhhvvhga]

Naturally occurring differences are necessary if you want to embed the watermark, otherwise detecting it is trivial - you just need to compare two copies. A photo or a scan of a text document can easily weigh 2 MB, that's enough data to embed a watermark in a way that is reasonably difficult to detect. Embedding watermarks that will survive re-encoding while being hard to detect is far more difficult though.