I wrote a few papers about this and have covered it a few times. The main thing holding back this type of vision substitution is the reliability and size of microelectrode arrays. You can get a 1080p camera inside the eyeball, but you have to put that information through the normal process if you're going to take advantage of the brain's vision systems. The earlier you put it in the stream the better, and retinal cells are handily arrayed in a predictable spatial pattern and the mechanisms of the cascades are fairly well understood compared with later downstream stuff that occurs in the V areas of the brain.

Stimulating the optic nerve directly is not really an option (too small, too fine, too unpredictable) and much further in the process and you start skipping critical parts of vision processing. That's why work on cortical blindness will continue for many years after retinal blindness and the like have been addressed.

That said, this stuff was far more experimental 10 years ago and now there are probably a dozen serious research labs doing real life trials of these things, but with different takes on power, input, encapsulation, and so on. It really is exciting but these nonreactive, long-term microelectrode arrays need to be improved if we're to get better resolution and finer control over things like color and shading.