Three coordinated changes:
1. nightPaletteNew.cloud: void -> raised (slightly lighter than
sky) so clouds have contrast against the new-moon sky
2. nightPaletteFull.cloudShadow: panel -> void so cloud undersides
fade into the night sky, while the lit topsides (border) stand
out clearly
3. getNightColor() now keeps 'light' (moonlight) bright and only
partially darkens sky/bg. The previous version squashed
everything toward black, killing cloud structure.
Shader side:
- max(0.15, uDayProgress) keeps a baseline moonlight tint on the
lit side of clouds even when uDayProgress happens to be 0
- The final darken pass is now driven by nightFactor (sunAlt-based)
instead of the unreliable uDayProgress
Inserted two extra palette entries around the sunset transition:
- alt -0.262 (-15deg): already full-night (void), so ~1h after
sunset the sky is properly dark instead of still 11% into the
long -5.7deg -> -90deg void ramp
- alt -0.087 (-5deg): dusk anchor (panel) — was at -0.1
- alt 0.087 (+5deg): morning anchor (blood) — was at 0.15
Net effect: +5deg to -15deg transition (≈80 minutes of real time
in summer) now sweeps through golden -> rust -> panel -> void
instead of golden -> rust -> panel -> panel/void.
Three fixes:
1. getColor() switched from 'sunPos.altitude > 0' to
'> sunPalettes[0].alt' (-pi/2). The day palette interpolates
through twilight to deep-below-horizon, so an hour after sunset
(sunAlt ~ -15deg) we now stay on the day palette and get the
intended dark 'panel' colour instead of switching to the moon
palette immediately.
2. getNightColor() now actually applies MIN_NIGHT_BRIGHTNESS via
shiftColor toward black (the old 'brightness' var was unused
dead code).
3. applyWeatherTheme() cuts the weather tint by up to 70% at
night, so a 'cloudy' theme no longer washes the night sky with
the brighter weather sky colour.
The cloud horizon flickered when the camera was far from the
deck because the regular raymarch steps aligned with the cloud
boundary. Two cheap fixes:
- Per-pixel jitter (hash21 of gl_FragCoord + uTime) shifts each
ray's start by up to one stepSize; the brain averages the
random offsets into a smooth edge
- Wider smoothstep (0.18 -> 0.24) softens the binary cloud/no-cloud
boundary the jitter is now dithering across
- Cruise altitude 2.6 -> 2.0 (just above cloud ceiling y=1.5,
gives the 'just above the clouds' airliner feel)
- Expose __currentSunAlt (radians) from animate() so the
renderer can tell night from day without misreading
realDayProgress (which mixes pre-dawn + post-sunset hours)
- Replace (1 - uDayProgress) star gate by smoothstep over uSunAlt:
nightFactor = 1 when sun is well below horizon, 0 when above
- Widen sky band so stars are visible from the horizon upward
(cockpit-perspective frame is mostly below rd.y=0.3)
- Cloud ceiling lowered from y=1.8 to y=1.5 (thinner deck)
- Cruise raised from y=2.0 to y=2.6 (clear margin above clouds)
- Dive endpoint raised from y=0.9 to y=1.2 (skims top, never deep)
- Dive exponent 5 -> 8 (much rarer dives)
- Early-out and ceiling bounds updated to match new y=1.5
- Cruise altitude lowered to y=2.0 (just above cloud top at y=1.8)
- Look direction nearly level (rd.y ~ -0.05 at horizon line)
- Stars rendered as 3D hash-grid sphere, only above horizon and
only when uDayProgress < 1 (i.e. at night). Subtle sin twinkle
keeps them from looking like static dither.
Camera altitude is now driven by pow(sin,5) so the view stays at
y=2.6 most of the time and only briefly dips to y=0.9 to skim the
top of the cloud deck. Marching also skips the empty gap between
camera and cloud ceiling when looking down from above.
Reduce raymarch iterations from 28 to 18 and fBm octaves from
4 to 3 (with step size 0.55->0.85 to preserve cloud coverage).
Per-pixel cost drops by ~55% while visual density stays similar.
Raymarches a horizontal cloud layer (28 steps, 4-octave 3D fBm).
Camera path is sinusoidal in altitude so the view dips into the
cloud deck at intervals for a 'fly through the edges' feel.
Density threshold from weather: sunny -> sparse, stormy -> overcast.
Renderer is selected via CONFIG.renderer ('clouds2d' or 'clouds3d').
- Remove all hardcoded fallback hex/rgb values
- Skip draw until __currentPalette is published by parent
- applyWeatherTheme now also mixes cloudShadow (was dropped)
The 2D WebGL renderer now creates its own <canvas> overlay inside
the target element, instead of expecting the target to already be a
canvas. This fixes 'canvas.getContext is not a function' when the
target is a <div>.
Richard Zink