The Visual Cognitive Load Index (VCLI-G)

“Failure ≠ collapse” → Aesthetics and Collapse Exposes Defaults, Choice, Coordinates and the Delta from the Center.

The Visual Cognitive Load Index

VCLI-G quantifies geometric compositional complexity through four measurable channels: centroid drift, void topology, curvature torque, and occlusion entropy. It measures geometry, not taste, showing whether complexity feels intentional or chaotic. Think of it as a diagnostic for how an image holds, not just how it looks. Paired with SCI (Structural Coherence Index), it provides a two-axis framework for analyzing compositional states, from Default Simple through Resolved Clarity, Chaotic Complexity, to Earned Tension. It estimates geometric complexity: how effectively structure sustains without collapsing into noise. The system offers practitioners navigational coordinates for steering generative AI systems toward specific structural outcomes. — See Github for code

Scoring is a Dial, Not a Verdict (VCLI-G) Profiles Explained

This document explains how scoring profiles act as adjustable lenses, not rigid judgments. Each bias setting, AI Conservative, Neutral, and Balanced Plus, shifts what the system rewards, from structural discipline to expressive tension. The goal is to treat the dial as a tool for intentional analysis: a way to align measurement with artistic aim, letting nuance survive where a single score would flatten it.

Reading the VCLI-G - Start from the Center

This is a single-case demonstration showing how VCLI-G tracks compositional decisions across systematic variations of one subject (the balloon house). It walks through eleven iterations, from centered baseline (emblem risk) through separated elements (friction), added material consequence (earned complexity), intentional collapse, and final synthesis - showing how geometric signals (G1-G4) respond to structural changes. The study maps each image's position in the VCLI-G × SCI phase space, revealing how tension builds, stabilizes, or collapses based on compositional choices. This is explicitly a tutorial document demonstrating measurement behavior, not a validation study claiming generalizability. It shows what VCLI-G measures and how to interpret the coordinates, using one concrete narrative arc to make the abstract framework legible.

A Geometric Framework for Compositional Analysis in Generative AI

This study documents a geometric measurement framework for analyzing compositional structure in images. Using a controlled cup sequence, it demonstrates that two independent metrics, VCLI-G (geometric complexity) and SCI (structural coherence), can reliably differentiate four compositional states across systematic variations. The work provides measurement infrastructure and navigational vocabulary for practitioners steering generative AI systems, with findings consistent across multiple subject domains.

Addendum A: Non-Linear Transformations for the VCLI-G Space

The addendum defines and formalizes how non-linear transformations extend the Visual Cognitive Load Index (VCLI-G) system into a perceptually normalized, distribution-aware model. It introduces two operators: Anisotropic Power Equalization (APE) and Residual Contrast Expansion (RCE), which reshape the canonical VCLI-G × Structural Coherence Index (SCI) plane. APE applies anisotropic power-law scaling to equalize perceptual sensitivity across axes, while RCE amplifies residual variance within coherence bands to expose fine-grained differences in artistic strategy. Together they convert statistical distance into perceptual distance, distinguishing geometric sophistication, perceptual robustness, and compositional intent. The document also explains why these transforms are context-sensitive (batch-dependent), how to stabilize them for reproducibility, and how their behavior mirrors human perceptual recalibration.