It is interesting to me that text to image AIs are not good at geometry.
They relate to geometry visually but not conceptually.

It is a while since I have tested out Midjourney or Dalle's ability to make geometries. 

When I last tried, they could make cubes, spheres, and not bad at icosahedrons (this is one of my better ones here),
but I never managed to get a good dodecahedron, tetrahedron, octahedron or cuboctahedron.

Of course we can cheat by giving it a reference image,
but has anyone managed to get any of these forms using pure text-to-image?
Please post examples if you have.