As described in Dercuano drawings, I want to add illustrations to Dercuano. Some of the cases where illustrations will help most are the shapes of three-dimensional objects, and it occurs to me that in many such cases it might be easier and quicker to specify the three-dimensional geometry of the objects than to sketch them by hand.
The trouble is that I want to make sort of casual sketches, and the semi-photographic quality of normal rendering makes conspicuous any insufficiency in the models being rendered; also, shiny eye-catching rendered graphics might be eye-catching enough to detract from the kind of thoughtful consideration I’d like the reader to be able to apply. So some kind of non-photorealistic rendering might work better.
In particular, I was thinking that maybe by using lines of varying width or varying darkness, I could get a kind of engraving effect, though maybe that’s too skeuomorphic. To the extent that the lines run in straight lines (geodesics) along the surface of the 3-D model, they can additionally help by showing the curvature of the surface; to the extent that their orientations aren’t determined by the view, the projection will tend to skew them in a way that shows the orientation of the surface even when it isn’t flat.
It might be simplest, and perhaps adequate, just to render things in grayscale, though. That might be adequately calm.
Slow, perhaps periodic rotation or nutation might also help with showing 3-D structure; perhaps motion-blurring it would prevent the motion from being too distracting.
A separate question is how to carry out the construction of the 3-D model. Of course this can be almost arbitrarily easy or arbitrarily difficult, but the methods available to date leave a lot to be desired. Listing center-coordinates, radii, and colors of spheres is simple, but there’s only so much you can build with spheres, and it’s a time-consuming way to build it. CSG is intuitive, for what it can express, but algorithmically it can be very challenging, and it’s worthless for modeling tree bark, filleted joints, or smooth curves, and again, it’s (often) very time-consuming to use. Teddy 3D is intuitive but difficult to achieve precise results with. The constraint-solving pipeline approach used by FreeCAD, CATIA, and SolidWorks (modeled on the approach taken by Sutherland’s SKETCHPAD) is clearly capable of constructing very complex shapes, but these programs’ solvers not infrequently fail to converge, and when they almost fail to converge, are quite slow.
Another thing to keep in mind is the possibility of printing. Black and white polygons print very well on laser printers, as do solid lines and polygons of the printer’s primaries if it’s a color printer; color and grayscale regions require some kind of halftoning, which dramatically reduces the resolution.