Geometry to Points

Every shape is generated in a normalized space (half-extent or outer radius = 1), then scaled non-uniformly by Size X / Size Y / Size Z, rotated (X → Y → Z, in degrees), and finally translated by Position X / Y / Z. With the default Size of 0.8, a shape spans roughly ±0.8 Grafit units — comfortably inside the [-1, 1] shortest-edge viewport range.

Sampling is fully deterministic: a seeded xorshift RNG means the same Shape, Distribution, Count, Seed, and parameters always produce the identical cloud, frame after frame. Nothing jitters unless you change an input.

Shapes

Distributions

Shapes without a natural corner set (circle, ring, grid, sphere, torus, cylinder, cone, tube) fall back to random sampling when Vertices is selected.

A few sampling details worth knowing:

• Circle / Ring volume uses square-root radial sampling, so density is uniform by area — no clustering at the center.
• Torus derives its geometry from Inner Radius: tube radius t = (1 - inner) / 2, major radius R = (1 + inner) / 2, so the outer extent is always exactly 1 before scaling. Edges mode quantizes the tube angle into 6 discrete rings running around the torus.
• Sphere Edges draws a wireframe of 6 longitude lines and 5 latitude rings.
• Pyramid and Cone interpolate their cross-section from base to top; Taper is the top radius (0 = sharp point, 1 = straight prism/cylinder).
• Cylinder Surface distributes ~70% of points on the side wall and ~30% on the caps; Cone Surface puts ~80% on the slant.

After sampling, the transform runs in a fixed order: rotate around X, then Y, then Z (positive degrees = clockwise, per the Grafit convention), then translate. The rotation pivot is the shape's own origin.

• Shape (Int, default 2 = Circle) — selects the geometry from the table above. Saved projects store the raw index, so the list is append-only and stable.
• Distribution (Int, default 0 = Volume) — Volume, Surface, Edges, or Vertices. For flat 2D shapes, Volume and Surface are equivalent (both fill the face); Edges gives the outline.
• Count (Int, default 200) — number of sampled points, clamped to 0…5000. Hidden in the inspector when Distribution is Vertices, because the corner count is fixed by the shape.
• Size X / Size Y / Size Z (Float, default 0.8 each) — non-uniform scale per axis, applied to the normalized shape. Size 1.0 means the shape reaches ±1 (full shortest-edge height). Size Z is hidden for flat 2D shapes (Square, Rectangle, Circle, Triangle, Line, Grid, Ring, Plane) since they live at z = 0. Stretch a Sphere with unequal sizes to get an ellipsoid, a Cube into a slab, a Circle into an ellipse.
• Inner Radius (Float, default 0.5) — the hole, as a fraction of the outer radius (clamped 0…0.98). Only shown for Ring, Torus, and Tube. Small values make a fat donut / thick pipe; values near 1 make a thin band.
• Taper (Float, default 0) — top radius as a fraction of the base (clamped 0…1). Only shown for Pyramid and Cone. 0 is a sharp apex; anything above 0 produces a truncated frustum (and, in Vertices mode, switches the pyramid from 5 corners to 8).
• Seed (Int, default 0) — reseeds the deterministic RNG. Same seed, same cloud; change it to get a different (but equally stable) arrangement.
• Rotate X / Rotate Y / Rotate Z (Float, default 0) — rotation in degrees, applied in X → Y → Z order around the origin. Rotate Z spins in the screen plane; Rotate X/Y tilt 3D shapes into perspective.
• Position X / Position Y / Position Z (Float, default 0) — final translation in Grafit units. Position Z pushes the whole cloud toward (+) or away from (-) the camera for downstream 3D renderers.

The Points output carries the cloud; the Count output reports the actual emitted point count (useful for driving other parameters, or checking the fixed Vertices count).

• Classic plexus constellation — Geometry to Points (Sphere, Surface) → Point Animation (Curl) → Plexus → Canvas. The staple network-of-lines look, breathing with divergence-free noise.
• Wireframe 3D object — Geometry to Points (Cube or Torus, Edges) → Point Transform (animate Rotation Y with an LFO) → Points Visualize → Canvas. A spinning point-wireframe with zero geometry assets.
• Shape morphing — two Geometry to Points nodes (e.g. Circle and Triangle, Edges) → Point Morph (drive Morph with an LFO or Audio FFT band) → Plexus → Canvas. Seamless dissolve between silhouettes.
• Star field with structure — Geometry to Points (Grid) → Point Animation (Drift, small Amount) → Points Visualize (small Size, Additive) → Canvas. The 8 × 8 lattice keeps an underlying order while drift keeps it alive.
• Composite scenes — Geometry to Points (Ring) + Geometry to Points (Line, rotated 90°) → Points Merge → Plexus → Canvas. Build compound figures from primitives, each placed with its own Rotate/Position before merging.
• Audio-reactive geometry — Audio FFT → Math (remap a band to 0.4…1.2) → Geometry to Points Size X/Y → Points Visualize → Canvas. The shape pulses with the music while the point arrangement itself stays rock-stable thanks to the fixed Seed.

• Count is capped at 5000. If you need a denser cloud, chain a Point Density node after it (Subdivide or Nearest) — its ceiling is 50 000.
• Vertices mode is the cheapest possible cloud (2–8 points) and pairs beautifully with Plexus at large Max Distance: instant geometric line figures.
• The pyramid's normalized height runs z = 0 to z = 2 (not centered) — use Position Z ≈ -0.8 (with default Size) or Point Transform with Center enabled to pivot it around its middle.
• Sampling is deterministic per Seed, so animating Count grows/shrinks the cloud without reshuffling existing points — handy for reveal effects.
• All parameters re-sample the cloud when changed, which is cheap at these counts, but the node is not time-based: for continuous motion use Point Animation downstream instead of animating Seed (which teleports every point).
• Prefer transforming here (Rotate/Position inputs) when the node is the only source; reach for a separate Point Transform node when you need to re-orient a merged or animated result further down the chain.