We generally work out image first ourselves, but sometimes we need to fix it up afterwards... and of course the image just doesn't pan out. Or the character doesn't quite match our plan. Or something like that. We tend to have more images than we have characters, at least.

Sometimes, an image will just give us a character like that. Other times, the character has to be worked so that it fits an image.

We're not beyond changing a picture so it fits our concept - its not really difficult to tweak hair color, eye color, skin tone, etc. 

But sometimes, there just ain't an image that fits a character. That is when those textual descriptions you mentioned higher up here come in handy. It takes skill to do a textual character image that shows fully what a character is like, and even then there's no single answer the way a regular image tends to have.