Procedural modelling builds geometry by evaluating rules. The asset is the rule set; the mesh is its output.
The trade
- Cost: setting up the rules takes substantially longer than modelling the thing once.
- Payoff: unlimited variations, non-destructive iteration, and changes that propagate everywhere at once.
The break-even is roughly "will I need many of these, or will I change my mind about it repeatedly".
Where it wins clearly
- Architecture. Buildings from footprint, height and style parameters.
- Environments. Roads, terrain, scattering, vegetation distribution.
- Modular kits. Generating consistent tiling pieces from shared rules.
- Anything with regular variation. Fences, stairs, pipes, cables.
Where it loses
- Unique hero assets. One statue is faster sculpted.
- Highly art-directed shapes. Rules produce rule-shaped results, and fighting them to hit a specific silhouette is harder than modelling it.
Simple procedural tools you already have
You do not need a dedicated package to work procedurally. Modifier stacks (array, mirror, screw, curve), parametric primitives, and node-based geometry systems in mainstream applications cover a lot of this ground without the learning curve.
Against generation
Procedural is exact, repeatable and controllable; generation is fast, plausible and one-off. A scene often wants both: procedural for the structured, repeated content and generation for the organic, unique dressing.
The important shared property is that both produce meshes that still need the back half of the pipeline — clean topology, sensible UVs, and a material.