Identity geometry and role-conditioned physique

One identity solve, six deterministic inspection views

One recurring person owns exactly one large pure-white 3×2 identity Inspector generated in one built-in ImageGen call. The fixed cells are:

1. top row: full-body front, one role/script-relevant true side profile, full-body back;
2. bottom row: large portrait front, portrait left profile, portrait right profile.

All angles therefore share one identity and body solve. Validate the whole Inspector first; any drift, anatomy/proportion error, oily or plastic skin, smear, CG finish, copied/topographic/maze skin texture, lighting mismatch, garment change, insufficient native subject pixels, softness, or panel defect blocks this person immediately. A failed Inspector never becomes a crop or downstream reference.

After approval, build-reference-board preserves the fixed three-column/two-row role order, detects and records one unambiguous near-white horizontal gutter near canvas center, then crops six PNG views without resampling or repainting. It does not blindly cut the mathematical midpoint if ImageGen made the full-body row slightly taller. A missing/ambiguous gutter, content crossing it, or cropped hair/footwear fails loudly. The crops are inspection and UI-upload aids only; the Inspector remains the first and only identity authority. Full-body cells use normal perspective, natural camera height, complete feet, neutral posture, and no story light, graphics, labels, props, or scenery. New boards require at least 1536×1024 native pixels, 512 pixels per column, a 560-pixel full-body subject, and a 300-pixel portrait face; dimension-only enlargement does not count. The legacy four-column form remains readable for existing v1.6 projects.

Proportion signature

Measurements are normalized to standing_height_1.0; the system does not impose an ideal eight-head body. It records head/body, shoulder and hip width, torso, arm segments, thigh/shin, arm span, hand/foot, asymmetry, handedness, source projection, confidence, and tolerance.

Physique is a casting fact

Classifications include observed-specific, narrative-fit, aspirational lean/athletic/muscular, fashion model, dancer/performer, strength/power, endurance-trained, story-specific deconditioned, controlled soft-tissue transition, and custom.

Target-driven categories require at least two measurable targets and two rejection criteria. A body cannot be “fixed” with a low camera, wide-angle leg stretch, automatic slimming, or wardrobe that invents a different silhouette.

Fictional adult leads default to a naturally measured 4:6 visual leg/body balance. Fashion, dance, action, and performance roles may use a justified 3:7 impression. An unintended 5:5 silhouette is a hard failure across identity, wardrobe state, and narrative frames; a level camera and normal perspective must prove the body rather than a low angle or wide lens manufacturing it.

Age, disability, size, and health are not defects. The gate asks whether the result matches the role and user intent while maintaining anatomically coherent joints, support, garment separation, and continuity.

Multi-person projection

A two-or-more-person frame inherits each person's approved standing height and body geometry from that person's unified Inspector. Record one rectilinear camera projection, camera height and pitch, every subject mark and camera distance, the expected and observed projected-height relation with tolerance, crop-aware body landmarks, horizon and vanishing lines, occlusion order, and a shared ground surface or explicit elevation offset.

For subjects near the optical axis without severe lens distortion, use the sanity check projected height A / projected height B ≈ (approved height A / distance A) / (approved height B / distance B). A cropped foreground body is checked from visible head–shoulder–waist or waist–knee landmarks; the workflow must not invent its off-frame height. Long lenses compress the impression of distance but do not cancel inverse-distance scale. Wide lenses may exaggerate near/far scale but may not stretch one person's legs, shrink a head, widen an edge body, or change shoulder/hip structure.

Per-person scaling, floating or buried feet, incompatible body and head perspectives, contradictory ground contact, or hiding a proportion error with crop, shallow focus, or motion blur is a hard failure.

State changes

Age, weight, pregnancy, swelling, or muscle changes use a documented soft_tissue_delta with cause and first visible shot. Skeleton, identity geometry, and all undeclared features stay locked. Every state returns directly to the approved Inspector; state-to-state edit chains are forbidden.