OptionalboundsHalf-extent of the generation volume. Lines: bounds of the random
walk. Points/gsplats: cluster centers are drawn uniformly from
[-bounds, bounds]^3. Larger bounds → more on-screen spread.
Defaults to [-100, 100].
OptionalseedDeterministic seed so the same (type, count, seed, clusters)
produces the same scene byte-for-byte. Defaults to 1.
OptionalclustersNumber of gaussian blobs the points/gsplats samplers draw from (ignored by 'lines', which is a random walk). Defaults to DEFAULT_CLUSTERS.
OptionalblendingBlending mode for the injected node — consumed by the debug injector (not the generators). Defaults: 'additive' for lines (the historical bench contract), 'normal' for points/gsplats so the depth-sort subsystem engages.
OptionalwidthLines only: per-endpoint width written for every segment. Defaults to 1.0 (the historical bench contract). The thick perf scenarios raise this so segments are much WIDER than they are long — the worst case for footprint-area cost (#1352 gate G0/G1).
OptionalstepLines only: random-walk step size as a fraction of bounds.
Defaults to 0.01 (the historical bounds * 0.01). Lower it to
make segments shorter relative to their width.
OptionalturnLines only: maximum per-step turning angle in RADIANS. When set, the walk carries direction momentum and turns by at most roughly this much per step — a smooth streamline/trajectory proxy. Unset keeps the historical fully-random walk, whose consecutive segments turn ~90° on average (an adversarial case for any join-aware renderer, measured on the #1352 spike).
Number of elements (segments / points / splats) to generate.