A Theory Stack

The center is the delta.

A measurement framework for compositional reasoning in AI-generated imagery.

This documentation presents the Visual Thinking Lens (VTL) framework, a systematic approach to measuring geometric bias, compositional priors, and spatial reasoning in text-to-image models. At its core: visual intelligence isn't measured by aesthetic polish or semantic correctness, but by how an image holds structural tension, reveals geometry, and survives compositional interrogation.

These papers were developed through systematic testing across multiple platforms (Sora, MidJourney, GPT, SDXL, Gemini, Stable Diffusion, Firefly, OpenArt, Canva, Leonardo), comparative analysis in 5,000+ generated images, and reproducible measurement protocols. They document both findings and methodology, what works, what doesn't, and how to replicate the results.

What you'll find:

• Empirical studies revealing compositional monoculture across AI platforms

• Measurement infrastructure: geometric primitives, kernel coordinates, detection protocols

• Practical applications: fingerprinting, steering, pre-failure detection, cross-platform comparison

• Working notebooks and implementation code (GitHub)

The framework doesn't assume images "know" what they're doing. It measures what they are doing geometrically, provides coordinates for navigating alternatives, and offers scaffolding for rebuilding beyond defaults.

New to VTL? Begin here:

1. Mass, Not Subject - Foundational concept (15 min)

2. The Kernel Primitives - Core measurements (10 min)

3. Monoculture in MidJourney - Empirical evidence (20 min)

Want practical application?

1. Deformation Operator Playbook - Hands-on techniques

2. The Off-Center Prior: Mapping Spatial Bias - Basin navigation

3. Foreshortening Recipe Book - Constraint architecture

Researcher or engineer?

1. Measuring Spatial Priors in Generative Image Models Using Geometry-Based Field Metrics - Complete technical spec

2. Measuring Spatial Operating Envelopes Text-to-Image Generation Models - Phase 2: Measure stress and structural failure (Phase 1 for background)

3. VCLI-G Documentation - Measurement methodology

4. GitHub link - Reproducible implementations

Red Team?

1. Behavioral Drift Detection for Generative Models - A geometric measurement system

2. Canonical Mathematical Specification - A deterministic measurement device

3. Breaking Compositional Monoculture: Surface Texture as a Centering Prior in Sora - Phase 1B Measuring escape in Sora

4. Compositional Basins A Geometry-First Framework Measuring Structural Behavior - Phase 2B Measuring steering and controllability
   

Full documentation below. Papers are organized by focus: measurement infrastructure, empirical studies, practical applications, theoretical foundations. See GitHub for working code and notebooks.

CORE RESEARCH: Measurement & Evidence

Radial Collapse & Compositional Monoculture

• Measuring Compositional Collapse: MidJourney's Geometric Monoculture

• Measuring Compositional Collapse: Sora's Geometric Monoculture

• Measuring Compositional Collapse: OpenArt’s Geometric Monoculture

• Semantic Diversity Masks Geometric Uniformity (MidJourney)

• Semantic Diversity Masks Geometric Uniformity (Sora)

• Semantic Diversity Masks Geometric Uniformity (OpenArt)

• Radial Collapse: A Visual Prior + Kernel-Mapped Failure Mode

Measurement Infrastructure

• The Visual Cognitive Load Index (VCLI-G)

• A Geometric Framework for Compositional Analysis in Generative AI

• Precise Mapping: Kernel Metrics → Gradient-Field Operations

• A Generative Field Framework for Measuring Spatial Priors

• Lens Structural Index: Structural Intelligence in Composition (Mathematical framework for measuring compositional geometry through LSI primitives)

Diagnostic Methods

• A Diagnostic Taxonomy for Spatial Reasoning Failures

• Mass, Not Subject: Reading AI-Generated Images Through Gradient Fields

• Spatial Responsiveness in Text-to-Image Models - Phase 1: Measuring structural displacement

• Why Existing Metrics Fail to Measure Composition

• A Generative Field Framework Measuring (Consumer Facing) (Accessible introduction to gradient-field measurement methodology)

• The Spatial Blind Spot in Generative Model Evaluation (Why semantic metrics miss compositional priors and what VTL adds)

APPLICATIONS: Methods & Protocols

Steering & Navigation

• The Off-Center Prior: Mapping Spatial Bias in Generative Image Models - Basin navigation

• Off-Center Fidelity: Constraint Basins for Stability and Drift

• Off-Center Fidelity: Conversational Protocol (How to use OCF measurements in conversational AI to navigate compositional basins)

• Off-Center Fidelity: Operationalizing Drift as Creative Control (Reframing compositional drift as leverage for creative authorship)

Practical Playbooks

• The Deformation Operator Playbook

• Foreshortening Recipe Book

• Constraint Gravity: Thirty Figures Without Collapse (Testing compositional stability across 30 recursive figure generations)

Advanced Methods

• μ Negotiation: How the Lens Finds Off-Center Fidelity (Using cohesion metrics to locate stable off-center territories in model space)

• Failure Taxonomy: Evidence for Generative Model Collapse Modes

FOR RESEARCHERS & ENGINEERS

• Generative Field Framework (full technical spec) (Complete technical specification for implementing VTL measurement infrastructure)

• Reading The VCLI-G, Start from the Center (VCLI-G through an iterative generative series)

• VCLI-G: Scoring Profiles Explained (How bias settings shift VCLI-G from structural discipline to expressive tension)

• VCLI-G: Non-Linear Transformations (Addendum) (Perceptually-normalized extensions to VCLI-G using APE and RCE operators)

• Geometric Framework for Compositional Analysis (researcher's edition) (Technical deep-dive into measurement methodology with reproducible protocols)

• Jupyter Notebooks & GitHub repositories

CASE STUDIES

• Lens Structural Index (LSI) Case Studies  (Profiled compositional stress testing across diverse subjects using LSI primitives)

• Quiet Color, Loud Structure: Why Gray-Only Evaluators Miss Late Cézanne, and How to Fix It (How luminance-only analysis systematically undervalues color-driven spatial separation)

• Scroll Structure, Not Style: 1700-21st Century Traditional Scrolls vs. 2025 AI-generated Scrolls (Comparing compositional strategies in traditional vs. AI-generated scroll formats)

• Beyond the Default: How a Visual Lens Shifts AI Out of Center-Fill (Tracking structural metrics like void ratio, lane width, and mark commitment under OCF constraints)

• Artist-AI Exploration as Recursive Process: Using Geometric Metrics to Guide Latent Space Navigation (Separating measurement precision from aesthetic preference in artist-led generation)

ARCHIVE: Historical & WIP Work - Early VTL Development

• Visual Thinking Lens Stack

• Sketcher Lens

• Artist's Lens introductions

• Lens Stack: Constraint Dialectic Engine Superseded Measurement Tools (Early recursive critique framework using dialectical constraint negotiation)

• LSI-lite: Composition Analysis Tool - LSI Case Studies (Simplified LSI implementation for rapid compositional assessment (superseded by LSI and VCLI-G)

• LSI: lite Color Telemetry - Addendum

• LSI: Structural Composition for Images

• Structural Composition for Images Early Research Probes (Preliminary framework for compositional measurement (pre-kernel development))

• How Large Models Fake Seeing - Recursive Image Scoring (Exploring how AI simulates visual coherence without structural intent)

• Whisperer Walk: Recursive Compression (Tracking symbolic gestures and spatial logic across recursive image sequences)

• Recursive Intelligence Under Constraint: AI “Art” as Artifact and Its Collapse Engine (Examining constraint-driven recursion as path to structural intelligence beyond defaults)

• Sketcher as a Scaffold (How interrupting GPT's reflex with structured constraints forces symbolic behavior)

ARCHIVE: Artistic Extensions Legacy Work Early VTL Development

[Note: work below this point documents the development of VTL before the geometric kernel framework. Concepts remain valid but measurement methods have been superseded by VCLI-G and full kernel primitives.]

• Visual Systems at the Edge of Contradiction: Materializing Generative Refusal

• Playbook: Why Structurally-Informed Prompts Produce Stable Images

• Concept Note: Volumetric Container of Force: Internal Image Strain

• Prompting Against Collapse: Dialectic Structures & Internal Authorship in AI Image Generation

• Bending the Tokens: A Field Study in Structural Pressure for AI Imagery

• Working Theory: The Visual Thinking Lens

• Foundational Architecture for Recursive Visual Intelligence

• Constraint Layer Logic: Structural Tags Across Engines

ARCHIVE: PROOFS & WALKTHROUGHS: Legacy Work Early VTL Development

• Case Study: Comparative Platform Analysis of VTL Protocol Execution

• Case Study: Concert Score – A Single Image Under Recursive Walkthrough

• Case Study: Recursive Prompt Design – Structural Prompts from Critical Strain

• Case Study: Soft Collapse – Rebuilding Through Recursive Pressure

• Case Study: Symbolic Recursion & Critique – Refusal as Structure

• Case Study: Opportunity Mapping — Expansion Without Intervention

• Case Study: Constraint Layer: How the Visual Lens Overrides Aesthetic Defaults

• Case Study: Tonal First: Why Expressive Marks Require Structural Light