Cloud Clumping System — Field Guide

2026-06-28 · every knob, in motion, from many angles · real GPU (RTX 4070 Ti SUPER, D3D11) · captured live off the renderer. Each sweep plays low → high → low so you read the whole range in one loop.

This is the first of the modular cloud systems — clumping: the rules that decide, at every point in the sky, whether there's cloud there and what shape it takes. It's all built from 4 sizes of 3D "static" (cellular/Worley noise) carved by a coverage threshold, with detail nested into bigger shapes. Below: each dial we can now turn, shown in motion so we can feel what it does — and a few creative camera moves of the whole thing alive. Drive it yourself at the bottom.

1 · Coverage the master shape dial

Low packs cloud into fewer, denser, taller masses with clear sky between (dramatic cumulus). High spreads it thin and wide into a flat overcast blanket. (Counter-intuitive: "more coverage" = thinner, flatter — not bigger clouds.) Same sweep, three vantages:

coverage ground-up — **ground-up** — the player's view. Low = full dramatic masses overhead; high = thins out.

coverage above — **from above** — low = distinct cauliflower towers; high = continuous sheet.

coverage top-down — **top-down** — watch the islands of cloud merge as coverage rises.

2 · Edge softness crisp ↔ wispy

Low = tight, defined cauliflower edges (solid, sculptural). High = soft, hazy, dissolving wisps (atmospheric, stratus-like). This was Bonkahe's single hard-coded value — now a real dial.

softness ground-up — **ground-up** — crisp solid mass softening into haze.

softness above — **from above** — the edges fray and the form spreads as softness rises.

3 · Weather uniform blanket ↔ dramatic regions

How strongly clouds gather into weather systems across the sky. 0 = an even, wall-to-wall field. High = big dense masses with real clear-sky negative space — the strongest composition lever in the set (gives the sky a subject and breathing room).

weather ground-up — **ground-up** — flat fill resolving into a defined mass with open sky.

weather above — **from above** — the showcase: a uniform layer gathering into a dramatic cauliflower mass.

weather top-down — **top-down** — clear-sky gaps opening up between the gathered clouds.

4 · Wind-lean subtle — a fine-tuning tool

Leans cloud bases downwind (a faithful port of a control Bonkahe ships off by default). Honest note: it's a subtle reshaping, not a dramatic tilt — kept for later wind/dynamics work. Watch the lower edges shift:

wind lean above — **from above** — base shifts as the lean ramps 0 → strong. Subtle by design.

Side-on captures of the lean were dropped — from horizontal angles the clouds wash out (see the note below), so the lean wasn't readable there. That washout is its own finding.

5 · The whole system, alive creative camera moves

The integrated look (coverage 0.80 · weather 1.4 · soft 0.08) with the animation running, from a few cinematic moves — the kind of thing the player actually experiences.

pan left to right — **pan L→R** — sweeping across the dusk mass from the ground.

dolly through — **dolly in** — pushing toward the warm-cored tower.

look down then up — **tilt down→up** — craning the gaze up the face of the mass.

Known weakness (honest): the system reads strong ground-up and top-down, but goes thin and washed-out from the side and the far horizon — the clouds lose their silhouette horizontally. That's a separate axis from clumping (it's aerial-perspective + the vertical density profile), not something these dials fix. Part is realistic distance-haze; part is worth its own dedicated pass. Flagged, not yet addressed.

Drive it yourself

The live renderer, with the new — clumping — panel (Coverage · Edge-soft · Weather · Wind-lean) plus the five analytical lenses:

▶ Open the live lab

Try these straight away:

dramatic cumulus → &cov=0.78&weather=1.4 crisp cauliflower → &cov=0.80&soft=0.05 flat overcast → &cov=0.95

Verified against Bonkahe's primary source (noise = FastNoiseLite Cellular/Worley, large inverted into billows; our port is line-faithful) and measured live across 16 vantages. Method: qa/tools/_lab-gif.mjs (drives the live render via CDP, hides the HUD, sweeps each uniform or moves the camera per frame) + ffmpeg palette gifs. Frozen animation on the knob sweeps so only the dial changes; animation live on the camera moves. The CD's eye is the verdict.