Terrain Scatter

A showcase piece, not an example. Geometry Nodes sine-hill Mesh Grid with an Index-jittered Instance-on-Points scatter, realized cubes replaced by closed-form, glacier-worn boulders seated on sampled dirt Z, then the shipped pipeline: unique-cell UVs, Cycles high-to-low normal bake, LOD chain, convex collider, Unity glTF export.

GN still builds the hill, with two closed-form micro-relief terms on top of the sine, and the instance grid. After realize+slabify, whose rim is bevelled into the walls, cube islands are deleted and each centroid is reseated as a sized stone, sunk until the ground closes around it on every side, then clamped above the slab floor.

It asserts budget conformance of the generated result. It does not witness an API contract. "It rendered without error" is not a check.

Composes skills geometry-nodes-python, mesh-editing-and-bmesh, bake-high-to-low, depsgraph-and-evaluated-data, engine-export-presets, and snippets bake_normal_high_to_low.py, setup_bake_target_image.py, lod_chain.py / decimate_to_budget.py, convex_hull_collider.py, export_preset_unity.py (helpers copied, not imported as a package). Hygiene combinatorics match examples/mesh-hygiene-audit (copied, not imported).

Intended size: 1.80 m square hill tile, ~0.16 m sine amplitude, nine embedded stones from cobble to boulder; outer AABB 1.800 × 1.800 × 0.609 m.

Hero. The render path dresses the tile as a hand-made meadow diorama. None of it is part of the asset, its budgets or its export; the dressing is drawn with a fixed seed, parented to the tile, and the asset sheet judges TerrainLow alone.

• Soil section. Render-only point attributes give each dirt vertex its depth below the hill surface over it (exact on the vertical walls, which are single quads). The cut sides layer down from the hill's own top: a turf lip, dark humus with roots, then level horizons in object Z on wobbled boundaries: banded ochre clay under a paler loam with the odd stone, a cobble bed, and jointed grey bedrock at the foot.
• Boulders. Weathered granite: a grey per stone, three scales of noise, feldspar and mica speckle, rain streaks on the flanks, crustose lichen blotches (grey-green and sulphur, sparse orange) on the up-facing stone, moss and a soil stain where it meets the ground, read from a render-only height-above-ground attribute.
• Ground cover. About 37 000 curved, tapering blades in clumps of varied height and hue (a few dry straw blades among the greens); taller grass and seed stalks ring each stone, where grazing and mowing miss in a real field; a turf lip is combed out and down over the cut edge; drifts of ox-eye daisy, buttercup, knapweed and clover; three shrubs (juniper, hazel, flowering gorse) placed where they clear the stones, the path and each other.
• Worn path. A bare-earth foot path with grit meanders across the hill. Its closed-form centre line is chosen at render time from a fixed candidate set as the one that threads furthest from every stone; the soil shader and the dressing share it, so no blade grows on it and the pebbles gather along it.
• Light. A warm late-afternoon key from camera-left throws the boulders' and the grass's shadows across the slope, a cool fill keeps every stone face off black, a back rim lifts the grass tips and the lip, and the warm wedge lands on the backdrop.

Why the stones are worn boulders now#

The third pass found the stones read as placeholders at 100%: faceted grey icosphere blobs, one with a pure-black flat face. Glacial erratics, the boulders strewn across upland meadows, are sub-rounded to sub-angular blocks of granite or gneiss, rounded by transport in the ice with a few flat fracture and abrasion faces, lichen on their tops and moss where they meet the turf.

• Shape. Each stone is a quad cube-sphere (4 segments a side for the boulders, size factor ≥ 0.9; 3 for the cobbles) pushed out to a superellipsoid of exponent BLOCK_EXP = 2.6, which squares the shoulders into a block, with two closed-form low-frequency lumps of up to LUMP_AMP = 13% and four shallow cleaves (CLEAVE_DEPTH 0.68). STONE_R_BOUND grows by the superellipsoid's plan bulge and the lump amplitude, so the relaxation still spaces the stones by their real reach.
• Shading. Smooth over the worn body, with edges folding more than STONE_SHARP_DEG = 48° kept sharp, so the fracture faces keep a crisp arris. Flat-shaded, the same stones read as low-poly placeholders.

Budgets#

Declared as named constants; every gate recomputes from the mesh, materials, UVs, evaluated LOD, collider, or export file.

Base triangles rose from 1194 to 1758 in the first quality pass: 9-vert hill became a 21-vert grid, and bevelled cubes became subdiv-2 icospheres. Outer Z dropped from 0.655 m (crate corners swinging through the slab) to 0.552 m of seated stone on the hill.

The second pass took them from 1758 to 2398: the four-segment rim bevel adds 640. The outer Z moved from 0.584 m to 0.570 m, because the stones now sink into the ground rather than perch on it, and the declared size was re-fitted to match. The collider fell from 72 to 36 triangles because it now hulls a coarse, unbevelled 9-vertex tile (COLLIDER_GRID): the chamfer and micro-relief are visual, and on the full-resolution tile they pushed the hull to 178, over its ceiling.

The third pass took them from 2398 to 2986, and the declared band was re-fitted from 1400–2800 to 2000–3800 around the new count, the same relative width. The 80-triangle icosphere stones became quad cube-spheres: 192 triangles for each of the four boulders, 108 for each of the five cobbles (1308 for the stones, against 720). At 80 triangles a 0.4 m boulder's silhouette showed its facets, and smooth shading cannot hide a polygonal outline. The outer Z rose from 0.570 m to 0.609 m: the blocky superellipsoid keeps more of its height than the cleaved ellipsoid did, and the declared size was re-centred on the measured 0.6085 m (the ± 0.015 tolerance is unchanged). The collider rose from 36 to 54 triangles, still under its ceiling of 120, because the rounder stones put a few more vertices on the hull.

DECIMATE COLLAPSE triangle counts are not guaranteed identical across series — the gate is a ratio band, not an exact count. This mesh matched on 4.5.11 / 5.1.2 / 5.2.1. Bake pixels are stochastic; the gate is has_data plus operator FINISHED, not byte-identity. Construction uses no RNG: jitter, cleave planes and relaxation are all closed form or fixed iteration. Export byte counts differ by 16 B on 5.2.1 (glTF serializer), not a gated axis.

Hygiene#

Recomputed from the generated mesh, not asserted about the script.

Why the stones are cleaved, relaxed and faceted#

The committed stones were smooth-shaded, near-white ellipsoids with a gentle sine bump: eggs or marshmallows on a pale clay slab. Two pairs also interpenetrated, because the GN scatter jitters by up to 0.22 m on a 0.575 m grid, and no budget compared stone to stone.

• Cleaved. Each ellipsoid is cut by N_CLEAVES planes whose normals and depths come from the stone's own centre, closed form. Every vertex beyond a plane is projected onto it, which leaves flat broken faces. The first pass shaded them flat, because smooth shading turned the cleaved ellipsoids back into eggs; the third pass replaced them with blocky superellipsoids that hold their shape smooth-shaded (above).
• Scaled. Cleaving takes about a third off each stone, and at the old size they read as pebbles, so every semi-axis is ROCK_SCALE = 1.35×.
• Relaxed. Stone centres are pushed apart to 2 × STONE_R_BOUND + STONE_CLEAR over a fixed number of symmetric passes, and clamped so each stone's bound stays on the tile. STONE_R_BOUND is derived from ROCK_SCALE and the largest semi-axis, bump and tilt, so scaling the stones widens the spacing with them. With today's numbers the unrelaxed scatter would also clear; the relaxation is what keeps that true when the scatter or the sizes change.
• Materials. Dark mottled soil and grey weathered stone, with roughness and a small bump from fine object-space noise.

Why the stones are sunk, sized, and the rim is rounded#

The second pass inspected the tile in clay, wire, six orthos and ground-contact close-ups, and found the tile read as a slice of cake with pebbles dropped on it.

• Sealed. Stones were seated by sampling the dirt once, at the stone's centre, and biting the lowest point 35 mm below it. On the hill's slopes that left daylight under the downhill side: measured around each stone in 8 azimuth sectors, no stone had a buried vertex in more than 6, two in only 4. Now every vertex samples the ground under itself (a ray down onto the dirt, captured before any stone is added), and each stone sinks until every sector's most-buried vertex is EMBED_SINK below it.
• Sized. All nine stones were within about 20% of one size, on a relaxed 3×3 grid, and the top view read as planted. STONE_SIZES gives each scatter point a factor from 0.55 to 1.45, boulders among cobbles, and the relaxation spaces each pair by the sum of their own bounds.
• Rounded rim. The hill met the vertical walls at a knife edge all round. The rim is bevelled in four segments, and the budget bounds the step in normal elevation between adjacent faces. Elevation, not dihedral: at the corners the chamfer strips meet at a right angle in plan, which is a rounded corner.
• Ground, not a sine. Two closed-form relief terms, 14 mm and 7 mm, ride on the hill in the same GN tree. The hill and rim are smooth-shaded, because flat shading printed the 21×21 grid on the ground; the walls stay flat (the stones were flat until the third pass).
• Materials. Stones are mid-grey (they were near-white where the key hit them), and the dirt darkens toward a subsoil colour below the slab lift, in object space, without adding a material.

stone_overlap_audit BVH-tests every stone pair. --pile-rocks skips the relaxation and draws the scatter in to 40% so neighbours collide: 8 pairs interpenetrate (7 with the second pass's stones), and the piece exits 20. Skipping the relaxation alone proved nothing in the second pass, because the cleaved stones missed each other unrelaxed.

Falsifiers#

Each violates one named budget. All were run on 4.5.11, 5.1.2 and 5.2.1 and returned the same code on each.

Run#

blender --background --python terrain_scatter.py --
blender --background --python terrain_scatter.py -- --skip-decimate
blender --background --python terrain_scatter.py -- --stray-vert
blender --background --python terrain_scatter.py -- --lift-z
blender --background --python terrain_scatter.py -- --poke-rock
blender --background --python terrain_scatter.py -- --float-rocks
blender --background --python terrain_scatter.py -- --box-rocks
blender --background --python terrain_scatter.py -- --pile-rocks
blender --background --python terrain_scatter.py -- --perch-rocks
blender --background --python terrain_scatter.py -- --uniform-rocks
blender --background --python terrain_scatter.py -- --sharp-rim
blender --background --python terrain_scatter.py -- --output terrain.png

Smoke passes no flags.

Exit codes#

File-local. 9 is a valid check code. 10 is reserved for gallery_framing.check_framing on the --output path. 15–19 are the hygiene and joint-fit family.