pouët.net

Yes, the torso is cunningly packed spline data with gradients, rendered in a way that achieves the same style of interpolation as diffusion curves. It's more of an explicit rendering than a sparse linear system solver, though, but the end result is pretty much the same. There's also some layered composition going on.

The spheres and plant/tentacle things are just simple procedural junk to fill up the last kilobyte. I could have used the space for antialiasing, but birds.

loady: It's not a gradient mesh, no. It's a lot more more like diffusion curves. Just not quite the same approach. I only skimmed it, mind, but in the paper farfar linked to it looks like the diffusion is done by solving a discrete Poisson equation where pixels close to the curve make up prescribed nodes. I used heavily quantised b-splines as the base and a less elegant ray-marching-ish method for the diffusion.

But I am looking at doing something new with gradient meshes of a sort.