I think using a 3/4 and 2/3 view of the 3D models' heads instead of the full face view we have now would already make a big difference (maybe with a bit of background that's not to distracting); the faces would look much more alive. With every character's pose a bit different, the heads tilted in a different angle, and a bit more facial expression, the portraits could look much more vivid, while the characters are still easily recognizable. And it would be very easy to implement.
