"A picture speaks a thousand words"
Images are fundamentally more information dense than text. As plantuml renders images of different sizes based on the content, you're not going to be able to easily diff two images. If you have two images of exactly the same size, you can do things like "diff", which show up where pixels have a different colour, but these don't look great and will be very hard to do - in general.
In my opinion, if you are trying to communicate the change then you'll get the best results by manually highlighting the changes (either changing the colour of the elements, or just a hand drawn red circle!). If you're just trying to keep track of changes, stick with the text - that's why I like plantuml, as it's text and not a picture and is therefore a lot easier to track the changes.