I personally picked up both wxHaskell and reactive-banana at the same time and it was fine. In practice the two are almost orthogonal—reactive-banana completely supplants parts of wx and doesn't touch the rest at all, for the most part. (I recall some wx-specific quirks that leak out into reactive-banana code, but nothing too bad.)
For me, the examples included with reactive-banana-wx were probably the most useful learning aide. Once you figure out how the examples work, you can modify them and see the changes right away, which is enough of a stepping stone to figure the rest out from the API docs.
As far as superimposing text on an image, that should only take wxHaskell-specific code. You'll probably have to mess around with the layout, but the two widgets (text and image) should still be the same and so act the same from the perspective of reactive-banana.