This may sound like a silly idea, but I ask the same question for every open source game.
Have you ever thought of using
LDraw to generate all your models?
Advantages:
* Well-suited to grid-based games (element sizes change/grow in discrete steps).
* Huge existing library of part elements. Few elements would need to be constructed from scratch.
* Wide-spread familiarity and low learning curve should make it easy to attract new art contributors.
* Everything looks like Lego!
Disadvantages:
* An intelligent conversion utility does not exist. Hidden and coincident surfaces would need to be merged/trimmed, and additional poly-reduction steps might need to be taken (by default, LDraw models use *huge* numbers of polygons).
* The part elements are in a transition period where some parts have transitioned over to the
CC-AL license, while others have not. (See
here.) The intent is that *all* parts will eventually be released under this license. I'm also not exactly sure this license is GPL-compatible.
* Everything looks like Lego!
Here's an example of what one person was able to do in this regard. Notice that the images are in-game screenshots! Except that the hard part of the conversion was done by hand instead of using a dedicated tool.
Obviously, the lack of a converter is the big barrier. And, the license may not be compatible. However, the time it would take one intelligent coder to write such a converter would be much less than the time required to create large amounts of brand new content.
I realize the project might be too mature at this point to scrap all the existing content, but I thought I'd throw the idea out there.