Gallery Journal Knowledge System V3 Health Overview Historical Graphic Metaphysical Experiments Unrealized Temples Map About

Heavens V3: System Architecture

Multi-model generative system for creating abstract geometric art — seeking the typical form through a Generative Alphabet of primitives.

Each image is a perceptual instrument — a typical form so stable and inevitable that the viewer perceives past and future simultaneously, extracted from time. The system explores geometric primitives as carriers of psychophysical state.
1057
generations
0.2%
grid coverage
1371/616358
cells explored
0.31
avg best score
7926
pipeline logs

Pipeline

Each generation passes through 10 stages. Every stage is logged to the database.

DomeRepertoiredome_repertoire ScenePlotgpt-5.4-pro CreativeIntentgpt-5.4-pro SnapshotBuildergpt-5.4-pro SceneDirectorgpt-5.4-pro ImageGengpt-image-1.5 CriticCouncil3 models Reflectionclaude-4.6 MetaReflectionclaude-4.6 Knowledgegpt-5.4-pro

Critic Council

Three models independently evaluate each image. Consensus score = median of three. Disagreement flag (spread > 3) marks images for human review.

VoiceModelStrength
Structural AnalystGPT-5Prompt-result gap, composition analysis
Emotional IntelligenceClaude Opus 4.6Deep introspection, state transmission
Visual GroundingGemini 3.1 ProMultimodal analysis, perceptual truth

Score Weights

MetricWeightWhat it measures
State Transmission40%Would the experience HOLD someone who simply existed with the image?
Primitive Power25%Is the target primitive the dominant visual force?
Perceptual Truth20%Does this feel like being inside a dome?
Preset Fidelity10%Does the image follow its assigned preset constraints?
Novelty5%Has the agent explored new territory?

Exploration Grid

The system maps the space: 8 Experiences × 12 Primitives × 6 Approaches = 616358 cells. Epsilon-greedy selection: 70% explore (least-visited cells), 30% exploit (mutate best cells).

Axis 1: Experiences (8)

Cosmic Containment (79) Levitation (81) Crystallized Light (77) Prayerful Burning (79) Rational Infinity (84) Womb Return (80) Suspended Time (83) Technological Sublime (83)

Axis 2: Primitives (13)

Point / Singularity (46) Circle / Boundary (42) Ring / Annulus (49) Ray / Beam (47) Radial Star (51) Sector / Angular Division (47) Grid / Cell Matrix (50) Square in Circle (47) Axis / Spine (46) Density Gradient (49) Void / Saturation Contrast (46) Color Layer / Stratum (52) Spiral / Vortex (51)

Axis 3: Approaches (8)

Isolation Intensification Inversion Collision Dissolution Impossible Stratification Fragmentation

Models

StageModelUpgrade from V2
Image Generationgpt-image-1.5gpt-image-1 → 1.5 (Elo 1264)
Scene Plot & DirectorGPT-5.4 Progpt-4.1 → gpt-5.4-pro
Critic (3 voices)GPT-5 + Claude 4.6 + Gemini 3.1Single gpt-4.1 → 3-model council
ReflectionClaude Opus 4.6gpt-4.1 → Claude (deeper introspection)

Positive Prompt Design

All prompts use constructive framing: defining what to achieve rather than what to avoid. Every instruction states the target quality positively.

Full Pipeline Logging

Every stage of every generation is recorded in the pipeline_logs table: stage name, model used, input/output text (truncated to 5000 chars), duration, token counts, and the exploration grid cell assignment. This enables post-hoc analysis of which prompt formulations and cell assignments produce the strongest results.

← Архитектура Небес