brianbrady.dev

Three.js / semantic weather / browser AI

Mnemonic Weather Engine

A browser-only AI art prototype that turns a text prompt into a living 3D weather system of ideas. Concepts become glowing bodies, relationships become arcs, tensions become hot fronts, and the whole map can be remixed locally or generated through an OpenAI-compatible API.

View repo

System shape

  • Renders a navigable Three.js scene with WebGL, OrbitControls, fog, post-processing, and raycast inspection.
  • Uses structured semantic map data: concepts, weights, moods, tags, relationships, tensions, questions, and motion profiles.
  • Provides four generation modes: Map, Tensions, Ritual, and Strategy.
  • Includes export tools for JSON and canvas PNG capture.

Why it belongs

This is the experimental side of the same systems instinct: take vague input, force it into a legible structure, and make the structure visible enough to navigate. The aesthetic is loose. The machinery is deliberate.

Materials

  • HTML, CSS, and modern JavaScript.
  • Three.js renderer, scene graph, camera, fog, OrbitControls, EffectComposer, RenderPass, and UnrealBloomPass.
  • Data-driven concept graph with procedural local remixing.
  • Optional API-backed live mode for generated semantic maps.

Interaction model

  • Type or keep a prompt in the control panel.
  • Choose whether the system should produce a map, tensions, ritual structure, or strategy read.
  • Generate with saved API settings, or remix the built-in demo map locally.
  • Drag to orbit, scroll to zoom, and hover or click concepts to inspect the semantic node.

Prototype boundary

The app is intentionally light: no build step, no server, and Three.js loaded from a CDN import map. API settings can be saved locally for the demo, but the repo calls out the right production boundary: model calls should move behind a server proxy instead of storing keys in the browser.