The Fall — pane-coverage mask comparison
Mask-only Normal-printing experiment against the installed v7 text-contrast card. Artwork, lettering, backing/frame geometry, text-protection mask, and shader settings are unchanged. User will compare the material appearance; this candidate is not an approved final recipe.
In the harness, compare:
- Baseline:
the-fall-extraordinary-v7-normal-1000 - Candidate:
the-fall-extraordinary-v7-pane-coverage-normal-1000
Reload or reopen the Card selector to refresh the catalog. Keep lighting, material, finish strength, camera and rotation identical between cards. Foil and Holographic in motion will show whether the restored sky/path coverage is useful or excessive.
- Candidate mask
- Full mask comparison: baseline left, candidate right.
- Sky/path detail: same order.
- Extracted pane coverage: white receives coating.
- Extracted lead protection: white suppresses coating.
- Validation
- Install record
The old recipe multiplies pixel saturation and value, so muted clouds, gray stones and texture can turn into black coating holes. The candidate gives pane interiors a 62–90% coverage weight: gently smoothed saturation influences the weight, without cutting neutral glass out. Separate local dark-ridge extraction protects divisions using morphological closing, a darkness/contrast gate and a small coherence opening. The line analysis suppresses isolated etched-texture specks; it does not blur the complete finish mask across the glass divisions. Actual Normal overlay/glyph alpha suppresses coating beneath active backings and the frame.
This is extracted coverage, not hand-traced semantic pane geometry. Some dark bark/illustrative detail is still partly protected; it should be assessed in the user's comparison. The sky and path lose their broad accidental dark patches while retaining line detail. Pane coverage is deliberately gray rather than uniformly white. No additional runtime textures or rendering work are introduced.
The source canvas and authoring inputs are 2000×2800 with the existing origin and registration. Data exports at 1000×1400 and 500×700 use bilinear downsampling to avoid ringing. Runtime masks are opaque grayscale RGBA. The candidate runtime face and text mask are byte-identical copies of the currently installed baseline. The baseline's three runtime files remain unchanged. The separate candidate pair and text mask are installed in ../../card-harness/public/card-art/; the catalog has been refreshed with zero errors.
Recipe and inputs are reproducible through recipe.json, input hashes, and build-mask.py:
python3 build-mask.py
Rebuilding writes this proof's staged files; installation is recorded separately. Checks cover morphology against Pillow, a neutral-pane probe, a sustained dark-seam probe, rejection of an isolated dark speck, exact source/runtime preservation, active overlay/glyph suppression, all four corner values, dimensions and opaque grayscale encoding. Static mask inspection is separate from moving-light/user approval. Only Normal is exported for this comparison; the existing masks for other printings are preserved until the candidate is selected.