Sky Texture Sun Elevation

A runnable example that builds a World whose Background Color is driven by ShaderNodeTexSky, then proves sun_elevation is load-bearing with two tiny Cycles EXR zenith probes. The gallery still is a sky study: a red-granite gnomon obelisk on a paved plaza, lit by nothing but the sky, shown at both elevations side by side so the contract reads at thumbnail scale.

What it witnesses: the Sky Texture contract AI-generated lighting code most often gets wrong across the 4.5 LTS → 5.1 window.

• Sky must drive Background Color. A near-black Background Strength alone is not a sky — the check asserts the Sky → Background → World Output links.
• sky_type renamed. 4.5 LTS uses NISHITA; 5.1 replaced it with MULTIPLE_SCATTERING (Nishita's successor). Assigning NISHITA on 5.1 fails; the enum no longer lists it.
• dust_density → aerosol_density. 4.5 exposes dust_density; 5.1 raises AttributeError and ships aerosol_density instead. AI code still emits dust_density.
• sun_elevation brightens zenith. Two CPU Cycles OPEN_EXR probes with a straight-up camera, Background Strength 0.05 (unclipped), compare zenith luminance at 8° vs 55°. Measured rise 2.25x on 5.1.2 and 1.50x on 4.5.11 LTS (gate ≥ 1.25).

What each check catches on failure: broken Sky→Background link (exit 6, checked after the renders), wrong sky_type for the running version (exit 3), lost sun_elevation round-trip (exit 4), dust_density / NISHITA present on the wrong side of 5.0 (exit 5), a non-working sky (zenith floor, exit 7 — where --unlink-sky lands), and an elevation that does not brighten zenith (exit 8 — rise below 1.25).

Version witness: sky_type and the dust/aerosol rename are the divergence; sun_elevation itself is stable. Zenith rise differs by model (Nishita vs multiple scattering) but clears the same gate on both binaries.

Stage deviation#

The World carries a Nishita / multiple-scattering Sky Texture instead of the default near-black Background, and there are no studio lights: the contract is the sky, so the sky's sun disc is the key and its dome is the fill. The still is a diptych, 8° | 55°, split by a dark 12 px rule, with one Background Strength (0.035) for both panels so sun_elevation is the only variable. If it did not take, the two panels would render identical.

What reads at a glance: at 8° a deep navy sky with a dusk glow on the left horizon, the obelisk's sun face raked orange, the plaza nearly dark and the shadow running long out of frame; at 55° a bright blue sky, a sunlit plaza, and a short shadow ending on the bronze meridian scale inlaid in the paving. The ground runs to the horizon, so there is no void below it.

Layer 1 still holds: view_transform='Standard', designed materials (red granite, limestone, bronze, matte gilt, brick-pattern paving), a chosen 35 mm camera on a TRACK_TO aim, no helpers in frame. Standard does not compress highlights, so the render path fails with exit 12 if more than 0.05 % of the still clips to white. The committed still has zero clipped pixels. The pyramidion is matte gilt rather than metal: a metallic cap Fresnel-glinted the 55° sun to pure white at grazing angles, whatever its roughness. Mean luma is about 0.20, reported against the calibration set as information under this deviation.

The sky's sun_rotation is a compass bearing, measured clockwise from +Y. The first draft assumed the opposite convention and put the shadow on the wrong side. _sun_rotation_for_shadow holds the measured mapping.

Framing#

Each 640 × 720 panel is its own frame, and the gnomon is measured in it through examples/gallery_framing.py (exit 10). Both panels share geometry and camera, so one measurement covers both. On 5.2.1 the fill is x 0.566 / y 0.794 and the margins are left 0.100, right 0.334, bottom 0.044, top 0.161. The ground, plaza and meridian inlay are stage.

Run#

# Zenith-luminance correctness check (tiny Cycles CPU EXR probes) — the CI check:
blender --background --python sky_texture_sun_elevation.py --

# Falsifier: Sky→Background unlinked. Must exit non-zero (zenith probe: no sky).
blender --background --python sky_texture_sun_elevation.py -- --unlink-sky

# Also render the gallery diptych (Cycles; framing gate exit 10, clip gate exit 12):
blender --background --python sky_texture_sun_elevation.py -- --output sky.png

Exit codes#

Per-script sequential checks. 9 is a valid check code; there is no rule against it.

The blender-smoke workflow runs the check on Blender 5.2 LTS and 4.5 LTS (5.1 on the weekly cron, the needs-5.1 PR label, or manual dispatch). Smoke does not pass --output, so exits 10 and 12 are render-path only. Its catalog falsifier is --unlink-sky (expects exit 7).