The Gallery Opens: The /one-shot-scripts v4 Art Upgrade Plan
🗼 The plan: v4 expands art direction into eight dedicated wings — generative systems, typography, data viz, shaders, motion, illustration, 3D environments, and audio
🏛️ The structure: Each wing ships as a companion file under
scripts/visual/, loaded only when a task needs it — zero overhead for logic work🎨 The proof: Every section below contains a live piece built with the v4 rules. Hover. Click. Scroll
🗓️ The calendar: v4.0 in May · v4.1 in June · v4.2 in July · v4.3 in August · gallery fully open by Q3 2026
This post is both a roadmap and a gallery. The roadmap names the eight wings of v4. The gallery is what you are already walking through — each section below contains a working piece built with the rules the new wing will teach.
The museum analogy: v3.4 taught one-shot-scripts how to look at what it builds. v4 gives it a studio, a kiln, a printing press, a stage, a darkroom, and a concert hall. Same protocol, more mediums. Think of it like a general contractor who learned to see the house — now we're handing them the keys to the art department.
The Floor Plan
Eight wings. Each a self-contained script the orchestrator loads only when the task profile requires it. This is the full layout.
v3.4 established that visual output needs direction before code. v4 says each kind of visual output needs its own dialect — and the protocol should know which dialect applies.
Why Eight, Why Now
v3.4 proved one thing: a single "art direction" section across the phases was enough to stop disasters. It wasn't enough to produce beautiful work in every medium.
A particle system and a data chart and a book cover all need "visual direction first." But the rules for particle velocity are different from the rules for chart hierarchy, which are different from the rules for cover typography. v3.4 treated them as one topic. v4 gives each its own room.
v4: Medium-Specific Dialects
Task touches data → load scripts/visual/data-viz.md. Task touches audio-reactive → load scripts/visual/sound.md. The orchestrator picks the right dialect before Phase 2.
v3.4: One Dialect for Everything
"Establish palette, build in layers, screenshot each step" works for UI. It doesn't know what to do with a bar chart's colour encoding, or the difference between 60fps motion and 24fps cinematic pacing.
Wing I — Generative Systems
Particles, flow fields, agents, cellular automata, L-systems, noise. Work that is not authored frame-by-frame but defined by rules that run themselves.
The piece above is a two-species system — 260 green prey (align, avoid, cohere, flee) plus six orange predators (nearest-neighbour pursuit with trails). Same three classical flocking rules as v3.4's toy version, plus ecosystem dynamics. The behaviour emerges; the palette was fixed before the simulation loop was written.
Wing I's companion file will teach: particle budgets (mobile cap, desktop cap, off-screen culling), noise taxonomies (value, gradient, curl, fbm, worley), the difference between a flow field and a vector field, Euclidean-steering primitives (seek, flee, wander, arrive), and when to use a rule-based agent system vs. a pure particle system.
Core discipline: Generative art is ruled by its parameters. The craft is tuning parameters until the output has character. The protocol will force a parameter pass before the emit loop is written.
Wing II — Kinetic Typography
Letters as design, letters as motion, letters as image. One of the most common failures in v3.4 was "the text is readable but it has no voice." Wing II fixes that.
Wing II will cover: hierarchy (size, weight, colour, contrast), rhythm (tracking, leading, optical alignment), variable fonts (weight axes, width axes, slant), kinetic typography patterns (drop, stagger, mask, morph), and the sin most AI output commits — treating type as a delivery mechanism for words rather than a visual element in its own right.
Hierarchy
Three levels minimum: hero, body, meta. Size ratios around 1.4–1.8×. Never three type sizes trying to look equally important.
Rhythm
Tracking tightens as size grows. Line-height opens as line-length grows. Ragged-right for short copy, justified never.
Motion
Staggered reveals read like sentences. Mass animations read like noise. Delay offsets in the 60–120ms range feel human.
Voice
JetBrains Mono says "system." Playfair says "editorial." Space Grotesk says "startup." Pick the voice before picking the stack.
Wing III — Data as Sculpture
Charts that tell a story, not just plot numbers. Wing III is the biggest leap from v3.4 — data visualisation requires almost the opposite of UI design, and the protocol has to know.
The piece above is 48 bars around a centre, each updated every 60ms against a small pool of values. Colour encodes magnitude. Position encodes category. No legend is needed because the form is the legend.
Wing III's companion file will teach: chart selection by question-type ("is this a comparison, a change-over-time, a distribution, a composition, a relationship, or a flow?"), colour encoding rules (sequential vs. diverging vs. categorical), truth-in-axis rules (zero baselines, unbroken scales, accurate aspect ratios), and the "remove to reveal" discipline — every ink mark must justify itself.
Data-viz sin of v3.4: Rainbow colour scales for ordered data. Colour is ordered. Rainbow is not. v4 kills rainbow scales by default; sequential and diverging scales become opinions the rubric enforces.
Wing IV — Shaders & Fragments
Pixel-for-pixel GPU art. Noise functions, signed distance fields, fragment techniques that run 60fps on a laptop. This is where the aesthetic budget is smallest and the visual payoff is largest.
The piece above is a faux-shader written in 2D canvas — the same principles apply in real GLSL. Multi-octave noise, colour ramp lookup, cursor-driven domain distortion, and output clamped to a tight palette.
The v4 shader palette
Every wing ships with a palette gallery. Here's what Wing IV will ship with:
Seven stops. Luminance-ordered. Perceptually uniform steps. The companion file will ship fifteen of these — a cold palette, a warm palette, a duotone for documentary, a candy palette for games, a sepia for editorial, etc.
Wing V — Motion & Animation
Timing is a language. Linear motion says "machine." Ease-out says "gravity." Spring says "alive." v3.4 knew animation existed. v4 gives the protocol a vocabulary.
Each ball travels the same distance in the same time. The feel is the curve. Wing V teaches the reader — and the protocol — to pick the right curve for the right story.
| Signal | Curve | When |
|---|---|---|
| Stopping | Ease-out | Object settling, UI responding to input, modal opening |
| Departing | Ease-in | Element leaving the scene, dismissals, unmounts |
| Travelling | Ease-in-out | State-to-state transitions with no entry or exit emphasis |
| Playful | Overshoot | Notifications, gamified rewards, character moves |
| Mechanical | Linear / Steps | Progress indicators, type-on effects, deliberate coldness |
| Weighty | Expo / Spring | Drawer slide-ins, physics demos, cinematic reveals |
Wing VI — Illustration & SVG
The medium with the smallest footprint and the largest stylistic range. A 4KB SVG can look like Matisse, a medical diagram, or a 1970s airline poster. v4 gives the protocol a studio.
The piece above is entirely SVG — paths, gradients, a Gaussian-blur filter for the eye glow, and a scanning rectangle animated via JavaScript. 3KB compressed.
Wing VI will cover: geometric construction (the face was built from two arcs and a baseline), colour as accent vs. structure, SVG filter composition (glow, noise, displacement), and the difference between decoration and illustration — the former fills space, the latter has a subject.
Wing VII — 3D Environments
Scene composition is where v3.4 worked hardest and still lost. Getting geometry right is not the same as getting a world right. Wing VII is about world-building.
The piece above is a 40×40 heightmap generated from layered sine waves, perspective-projected to 2D canvas, orbiting slowly. No WebGL — the rules still apply.
Wing VII will teach: scene hierarchy (sky, horizon, mid, foreground), camera grammar (wide, medium, close, rule of thirds for 3D), lighting vocabulary (key, fill, rim, ambient), atmospheric perspective (depth via desaturation and fog), and why a single decisive light source beats five neutral ones.
Camera First
Pick the shot before you place a model. A bad model in a great shot reads better than a great model in a flat shot.
One Light Wins
Decisive direction — one dominant key light — beats three balanced lights. Shadows are where the drama lives.
Atmosphere Is Depth
Fog, haze, and colour shift do the work of geometry. Objects 100m away should be a different colour than objects 10m away.
Silhouettes First
Every hero object should read as a black shape on a white background. If the silhouette is mud, no amount of texturing saves it.
Wing VIII — Sound & Rhythm
The last wing and the most ambitious. Web Audio, audio-reactive visuals, procedural sound, and the vocabulary of rhythm itself.
The piece above simulates a frequency spectrum and renders it radially — 64 bars around a centre, pulsing to a compound rhythm, with a waveform ring at the outer radius. The same rendering code would drive real AudioContext.createAnalyser() data one-to-one.
Wing VIII will teach: Web Audio topology (sources → nodes → destination), frequency-domain vs. time-domain visualisation, beat detection basics, the prefers-reduced-motion contract for audio-reactive visuals, and — the hard one — procedural sound generation that doesn't feel computery.
The honest admission: this is the wing most likely to slip. Audio is full of hidden constraints (autoplay policies, mobile latency, battery drain, headphone vs. speaker mixes) that we'll discover by building with it. v4.3 is the target but the date is the softest on the roadmap.
The Palette Library
Every wing ships with its own palette set. Below is an abridged slice of the library — 12 palettes covering the most-used moods. Each one is a named constant the orchestrator can reach for. No task invents its own palette from scratch.
Why a library, not "generate a palette": Most AI-generated palettes fail one of three tests — the light end and the dark end have the same perceived luminance, two neighbours are indistinguishable to colour-blind viewers, or the accent does not pop at small sizes. Curated palettes pass all three before they're added to the library.
Technique Anatomy: A Shader, Dissected
A single piece, seven passes. Click a step to highlight what it adds. Each layer alone looks underwhelming. Stacked, the composition reads as intended — the v4 pipeline teaches the protocol to build this way.
Why this matters for the protocol: v3.4 told the builder "screenshot each layer." v4 teaches what the layers actually are. The shader companion file will contain this exact seven-step pipeline as a template, not a suggestion.
The Rubric, Diffed
v3.4 scored visual work on five dimensions. v4 adds four and reshapes one. Here's the full diff — green rows are additions, amber is a substantive rewrite.
prefers-reduced-motion honoured.
0.08
aria-label on canvas/svg, keyboard operability.
0.09
A Sample of the Companion File
Each wing's script lives alongside its phase scripts. Here's an illustrative excerpt from what scripts/visual/generative.md is shaping up to look like.
The Opening Schedule
Eight wings shipping across four releases. Each release is a working skill — nothing ships "half open." If v4.1 slips, v4.0 users still have three complete wings.
| Release | Ships | New Files | Breaking? |
|---|---|---|---|
| v4.0 | Generative, Typography, Data Viz | 3 companion scripts | No |
| v4.1 | Shaders, Motion | 2 companion scripts + palette library | No |
| v4.2 | Illustration, Environments | 2 companion scripts + SVG filter library | No |
| v4.3 | Sound & Rhythm | 1 companion script + audio policy notes | Possibly — audio gated behind opt-in |
The Permanent Collection
These are the v3.4 pieces — the gallery's founding works. Each one was built with a single dialect ("art direction"). Imagine what they look like once the protocol has eight dialects to choose from.
The Curator's Manifesto
Eight wings. Four releases. One protocol. The principles below are what every wing inherits — the guardrails the curator enforces before any piece leaves the studio.
📖 Rules of the House
- Direction before pixels. Palette, voice, reference, and mood board precede a single line of drawing code. No exceptions — not even for "quick" demos.
- Medium-specific dialects. Data viz follows data rules. Illustration follows composition rules. Audio follows Web Audio rules. The orchestrator loads the right dialect for the task.
- Screenshot every layer. v3.4's biggest win, inherited and enforced. Build a base → verify visually → add a layer → verify visually. Never "build it all then look."
- Performance is part of the art. A piece that drops frames is a piece that is broken. Mobile caps, reduced-motion fallbacks, and battery budgets live in Phase 4, not Phase 7.
- Accessibility is non-negotiable. WCAG AA contrast, keyboard operability, screen-reader labels on
<canvas>and<svg>, motion-reduction respected. Beauty does not override people. - Every wing ships with a palette library. Not one palette — a curated set. A task picks from the library; it does not invent its own.
- Honest about trade-offs. If a piece cut a corner to fit a budget, the delivery log says so. The rubric flags it. The curator does not pretend.
The Core Insight
Why this matters: v3.4 stopped one-shot-scripts from producing visual disasters. v4 is what lets it produce visual statements. The gap between "doesn't look broken" and "looks deliberate" is the gap between output and art — and it's the gap most AI tools never cross.
The museum analogy holds all the way down. A general contractor with blueprints can build a functioning house. A general contractor with blueprints and a gallerist and a sound engineer and a lighting designer and a typographer and a landscape architect can build a place.
That's what v4 is building toward. Not more capabilities bolted on — a set of dialects the protocol knows when to reach for. By Q3, every task with a visual or auditory component will know which room it belongs in, and bring the right tools before a line of code is written.
Get the Protocol Before the Gallery Opens
One-shot-scripts v3.5 ships today. v4.0 opens three wings in May. Everyone on the roster gets the upgrade the moment it lands.
Get Godmode Read More