Zetta P1 — procedural 3D, built on Claude

Zetta P1

Rather than reconstructing a shape from pixels, P1 reasons about structure and generates geometry as a set of operations. Built on top of Claude, and suited to anything with rules.

What it does

Structure, not pixels

P1 works from an understanding of how a thing is put together, so it can produce geometry that follows rules rather than approximating a photograph.

Repeatable variation

The same rules produce many consistent variants. Forty buildings that share a language and differ in detail, rather than forty unrelated meshes.

Dimensions that hold

Because the geometry is generated rather than inferred, measurements are stated rather than accidental — which is what modular kits and mechanical parts require.

Built on Claude

The reasoning layer is Claude, which is what lets P1 take a description of a system and produce the operations that build it.

How it works

  1. Describe the systemNot just the object — the rules. Grid size, module dimensions, how pieces meet, what varies and what stays fixed.
  2. P1 reasons about structureIt works out the operations that produce the geometry, rather than reconstructing a surface from an image.
  3. Generate and varyProduce one, or produce the set. Variants share their rules, so they fit together.
  4. RefineThe output is a mesh like any other and every refinement tool applies.

What people use it for

  • Modular environment kits whose pieces must tile and align exactly
  • Architecture from a footprint, a height and a style
  • Mechanical and hard-surface parts where a dimension matters
  • Any asset family where consistency across variants is the point
  • Content where you would otherwise be building a procedural graph by hand

What it cannot do

  • Weaker at organic and sculptural form — that is what N1 and N2 are for
  • A vague description produces a vague system; P1 rewards specificity about rules
  • One-off hero assets are usually faster to reconstruct or sculpt than to specify
  • Not a CAD tool: it generates meshes, not parametric surfaces for manufacture

The details

Approach
Procedural — geometry as operations
Built on
Claude
Strongest at
Structured, repeatable, dimensioned content
Weakest at
Organic and sculptural form
Output
GLB, with OBJ and FBX available