Scenario Zbrush Hard Surface

ZBrush hard surface

Hard surface is judged as a result: it reads as manufactured when every edge is clearly straight or clearly curved, planes carry an even highlight, edges are wide enough to catch light and every part could be built (Plouffe, Klimer). The agent builds that language from deterministic operations (primitives, Live Boolean stacks, polygroups as the crease plan, Dynamic Subdivision numbers, ArrayMesh, Panel Loops) and uses clicks and strokes only where nothing else exists. If a sibling skill named here is missing from your available skills, ask the user to install it (npx skills add scenario-labs/skills --skill <name>); unattended, proceed from tool schemas and flag the gap.

REQUIRED BACKGROUND: scenario-zbrush-expert (bridge, stroke engine, review loop, 2026 traps). Toolkit: scripts/zb_hardsurface.py (import sys; sys.path.insert(0, "<this skill>/scripts"); import zb_hardsurface as hs, then hs.hs_call("new_part", "cube")), built on the lead's zb_ops, zb_review, zb_audit and zb_stroke. Nothing in it has run in ZBrush yet.

Stance (the expert delta)

• Polygroups are the crease plan. Groups By Normals at 45 degrees (33 catches softer planes, 60 misses edges), Mirror And Weld, UnCreaseAll, Crease PG (Pavlovich; Plouffe). Crease first, a loop only where the falloff must be local. Crease PG after later edits recreases borders you softened on purpose: regroup, or crease by angle.
• The gap SmoothSubdiv minus CreaseLvl sets the edge. 0 reads razor, "very CG"; the closer CreaseLvl sits to SmoothSubdiv, the tighter the edge, so 4/3 is tighter than 4/2 (hs.edge_width). Pavlovich works at 3/2 and bakes 4/2; Plouffe keeps CreaseLvl 3 over a razor 4 and calls 2 too soft. Deciding condition: render both under metal, keep the widest edge that still looks planned.
• QGrid is for boxy blocks (it facets arcs). Curved spans take Smooth Subdiv plus Crease Tolerance, tuned per span: 22 held Pavlovich's arch; too low creases the arc's small angles.
• Booleans need clean inputs (Live Boolean doc; Pavlovich): watertight, no coplanar faces, similar density, low Dynamic levels, nothing partly hidden. DSDiv on, one Start group per separate part, then ZRemesher Detect Edges at Half. Creases, UVs, levels and layers die in the boolean; polygroups survive.
• Keep cuts and repeats live. ArrayMesh the cutter, not the result. Instanced fasteners (NanoMesh, ArrayMesh) make a late "flathead, not Phillips" one Edit Mesh swap; per-seat imports cost one call per seat.
• Straight or curved, never in between (Plouffe). Polish planes, then edges, then corner softness, sharp to smooth; never mix sharp and round edges close together; one cavity along a border sells a separate plate.
• Function and camera set the detail (Klimer): how it is made, where power flows, one fastener family, repeated angles identical. Validate twice in engine: the block (500 to 40k polys) with the rig, then a decimated high under 1M points. Unseen interiors and undersides stay filled or simple, sold by dark AO (Pavlovich).
• Micro detail waits for sign-off, on 3D layers (Klimer): a negative strength carves in, duplicating doubles, and a layer dies with any topology change, so layers come after the last boolean, remesh or Apply [added].
• Mirror And Weld always copies negative X onto positive X. Good half on +X: Deformation Mirror first (Pavlovich, Plouffe). Slice, Trim and BRadius ignore symmetry, so every one-sided edit ends here.

What the agent can and cannot do here

Establish first

Workflow

1. Setup. Proxy or concept as a SubTool; Unify, Z forward, X symmetry; versioned ZTL before every boolean, DynaMesh, ZRemesher and Apply (no undo from Python). GATE: longest bbox side about 2; ZTL on disk.
2. Block. new_part and place, or import_part from a DCC block; body first, cutters below. GATE: forms sheet (zb_review.review) against the concept; a rigged asset's block OBJ checked in engine with the rig before detailing.
3. Shell. Deformation sliders on primitives, subtractive boxes for openings, Make Boolean Mesh, zremesh_recipe("after_boolean"). The sculpted route (DynaMesh 128 to 176, hPolish, TrimDynamic) is concept only. GATE: hygiene(obj) ok (it stands in for Check Mesh, which reports through a note); PolyFrame shows even quads.
4. Crease plan. crease_by_groups(45), dynamic_subdiv(intent="working"); arcs crease_by_angle, tuned per span. Loops and insets made with ZModeler take their crease at creation (zmodeler_plan: Insert EdgeLoop "Crease"; Inset "Crease New Edges" crisp, "Crease Inner Poly" rounded; defaults are Do Not Crease). GATE: check_crease; metal sheet (review_sheets) shows soft, light-catching edges, no scalloping, symmetric polygroups.
5. Cuts and vents. Cutters new_part(..., op="sub") through the surface, never flush; Smooth 3 / CreaseLvl 2 plus crease by angle; array_mesh on the cutter. GATE before: stack_report()["problems"] empty, coplanar_report clean, Show Coplanar snapshot has no red, array_plan walls ok. Then make_boolean_mesh (short timeout). GATE after: UMesh SubTools equal Start groups; Show Issues snapshot clean; recrease (Crease PG unless a border was softened on purpose); hygiene before ZRemesher. Any one-sided edit: mirror_report, mirror_weld(keep=...), then mirror_check on exports before and after.
6. Panels and fitted parts. Plan panels as polygroups first. Real separations: Panel Loops with Plouffe's recipe (panel_loops). A visor or hatch that fits its opening: one cutter used twice (fitted_split_groups: Intersect in one Start group, Subtract in the other), or DynaMesh Groups on a group outline. Seams: a thin subtractive cutter from border_polylines plus curve_mesh [verify]. Masks: mask_fillet (Klimer's blur, invert, repeated). GATE: split_copy("groups") counts panels; gaps constant on the sheet.
7. Hardware. One family. ArrayMesh rows; radial sets as ArrayMesh Rotate 360 then array_commit("nano") with ZRVar variation (Chervenka); a part from an IMM without a drag: fastener_from_brush; no face where a bolt goes: a helper plane. A port across a seam: seam_port_groups (Pavlovich). Exact runs: spaced_points. GATE: count, orientation and the same fastener everywhere on the sheet.
8. High-poly commit. bake_split(): Apply at 4/2 on a copy, Dynamic off on the original. Exact chamfer widths: Crease Bevel width or a boolean chamfer, since Polish only gives the minimum (Polycount). GATE: levels = 1 + Flat + Smooth; edge_language has no ambiguous lines; plane_flatness no wobbly groups; pinch_report flags nothing.
9. Micro detail (after sign-off). Layers per region (Tool:Layers:New; procedures P15). GATE: layer_safe(recorded, zb_ops.stats()) before and after each pass.
10. Low poly and handoff. Dynamic-off cage; ZRemesher per part (zremesh_recipe("part")); quick static prop: dynamesh_visible(res) then one decimate_percent; or Maya or Blender. Klimer's ZRemesher plus auto-reduction gave "the nastiest mesh I've ever shipped". Painter routes (scenario-zbrush-retopology-export runs them): Substance Bridge Low & High plus Auto-Bake plus Send PolyPaint (bridge_preflight), or native Unwrap with crease seams (unwrap_creases) then OBJ. GATE: zb_audit.verdict(..., "game"); animated parts separate.