CAD describes shapes mathematically. Mesh modelling describes them as triangles. The distinction explains a lot of otherwise confusing behaviour.
CAD: exact surfaces
A CAD model stores a cylinder as a cylinder — an axis, a radius, a height. It is exact at any zoom level, editable by changing parameters, and carries manufacturing intent (this hole is 6mm, this fillet is 2mm).
Used for anything that gets manufactured: mechanical parts, product design, architecture, engineering.
Meshes: approximated surfaces
A mesh stores that cylinder as some number of flat faces approximating a curve. It is an approximation chosen at export time, it has no parameters, and editing it means moving vertices.
Used for anything that gets rendered in real time: games, film, visualisation, AR.
Converting
- CAD to mesh is straightforward and lossy in a controlled way. You choose a tessellation quality; higher means more triangles and a closer approximation. This is a normal, everyday operation.
- Mesh to CAD is hard and usually unsatisfying. Reconstructing parametric intent from triangles is inference, and the result is rarely as editable as a native CAD model.
Why CAD models make bad game assets directly
- Tessellation density is uniform, not distributed by visual importance. A hidden internal fillet gets the same density as a visible edge.
- No UVs. CAD has no texture layout concept.
- Topology is triangle soup, unsuitable for deformation or subdivision.
- Assembly structure is engineering structure, not rendering structure — often thousands of tiny separate parts.
A CAD model destined for real time needs the same retopology-and-unwrap treatment as any other high-poly source.
Where generation sits
Firmly on the mesh side. Generated output is triangles approximating a shape, with all the same properties — which is why a generated part is a poor starting point for anything that must be manufactured to a dimension.