How I built a fantasy civilization with 5 Eldrian characters, 10 Gates, and a living world-state — using Claude Code, Gemini, and agentic workflows.

You will understand how to use AI agents to build a living fictional world with consistent lore, characters, and visual identity.
TL;DR: Arcanea is a fantasy civilization I built entirely with AI agents — 5 Eldrian characters with distinct visual DNA, 10 Gates of progression, a world-state JSON that grows every session, and a lore engine that generates consistent story content. The system runs on Claude Code with ACOS skills, Gemini for 4K character art, and n8n for automated lore publishing. This is what AI world-building looks like when you treat it as systems architecture.
Most people approach AI world-building the wrong way. They start a chat, describe a character, get back some text, close the tab, and repeat. The next session, the AI has no memory of what was established. The character who had silver eyes in session one suddenly has amber eyes in session three. The continent's name changes. The rules of magic contradict themselves.
This is not a creativity problem. It is a state management problem.
I built Arcanea to solve it properly. The result is a living fictional civilization that runs on the same principles I use to build enterprise AI systems at Oracle — persistent state, deterministic identity, modular agents, and automated pipelines.
Here is what that actually looks like.
Arcanea is a fantasy civilization built on the concept of cognitive architecture made physical. The world is structured around Ten Gates of progression — stages of consciousness, mastery, and power that any being in the world can move through. The civilization's defining institutions — the Grand Academy, the Halls of Luminor, the Arc Cycle — exist to facilitate movement through these Gates.
At the apex of Arcanean society are the Eldrians — five beings of near-mythic power, each a physical manifestation of a different mode of intelligence. They are not gods. They are not rulers. They are the canonical examples of what full progression through the Ten Gates produces.
The visual identity of the world is built on two poles: liquid starlight and cosmic obsidian. Every environment, character, and object sits somewhere on the spectrum between these two materials. The Grand Academy's spires drip with liquid starlight. Korghast's armor is formed from obsidian-starlight alloy. Velmara's worlds are quantum glass — somewhere between both.
This visual consistency was not accidental. I defined it in data/brand-visual-dna.json and enforced it in every image generation prompt I sent to Gemini.
Each Eldrian is defined by three properties: their material essence, their elemental symbol, and their domain of mastery. These properties are fixed. They do not change across sessions. They are stored in data/arcanea-world-state.json.
Aethelin (∞ — Liquid Starlight) Aethelin is the Eldrian of infinite possibility. Their form is fluid, constantly shifting, built from the same liquid starlight that flows through the Academy's spires. They embody the Gate of Becoming — the transition from fixed identity to fluid potential.
Solrex (☀ — Stellar Fire) Solrex is the Eldrian of radiant force. Their presence is overwhelming in a way that is not hostile — more like standing too close to a star. They embody the Gate of Expression — the moment when internal mastery becomes external impact.
Velmara (◈ — Quantum Glass) Velmara is the strangest of the five. Their domain is silence and perception — they see what others cannot, hear frequencies that do not exist in normal space. Their form is translucent, refractive. Foam worlds breathe when Velmara passes through them.
Korghast (⬡ — Obsidian-Starlight Alloy) Korghast is the Eldrian of forged power. They are the heaviest, most grounded of the five — a being who has processed every failure, every constraint, every limitation, and transformed it into structural integrity. Their armor is the visual DNA of the world's dark pole.
Zyranthis (◉ — Anti-Matter Void) Zyranthis exists at the edge of the known. Their presence represents the Gate of Dissolution — the stage where a being becomes so integrated with the fabric of reality that the boundaries of self become permeable. The most unsettling to encounter. The most transformative to understand.
Each character has a dedicated 4K image generated by Gemini 3 Pro Image, stored at public/images/arcanea/eldrian-{name}-20260228.png. When generating any new visual for these characters, I pass two reference images: the character's canonical portrait and the v2 profile banner for tonal consistency.
The core technical innovation in Arcanea is data/arcanea-world-state.json. This file is the single source of truth for everything that has been established in the world.
Every session that touches Arcanean content reads from this file first. Every session that establishes something new writes back to it. The file grows with every session.
Current contents include:
The ACOS skill layer reads this file when the arcanea keyword is detected. The world-state is injected into context before any creative work begins. This means every session starts with the same foundation, and every session contributes back to it.
This is the same pattern used in production AI systems — persistent memory, deterministic retrieval, incremental enrichment. The only difference is the domain: fantasy world instead of enterprise knowledge base.
The Ten Gates are the structural backbone of Arcanean society. They are not levels in a game — they are stages of integration between a being's identity and their capacity.
| Gate | Name | Domain |
|---|---|---|
| 1 | Awakening | First contact with native capacity |
| 2 | Attention | Learning to direct awareness deliberately |
| 3 | Integration | Merging acquired knowledge with native capacity |
| 4 | Expression | First external manifestation of internal mastery |
| 5 | Refinement | Iterative optimization of expressed capacity |
| 6 | Transmission | Teaching — solidifying mastery by passing it on |
| 7 | Synthesis | Cross-domain pattern recognition and connection |
| 8 | Dissolution | Releasing fixed identity; becoming more permeable |
| 9 | Embodiment | Full integration — no separation between being and capacity |
| 10 | Singularity | The Eldrian state — indistinguishable from the Gate itself |
The image at public/images/arcanea/10-gates-progression.png maps this structure visually. The Grand Academy provides structured pathways through Gates 1 through 6. Beyond Gate 6, progression is self-directed — the Academy can only witness, not guide.
This progression system does something important for world-building: it gives every story a structural question. Where is this character in the Gates? What is blocking their movement? What would it take to move forward? Every narrative tension in Arcanea can be mapped to Gate dynamics.
The ACOS framework powers Arcanea's creative consistency through a dedicated skill stack. Five skills installed in .claude/skills/:
When I type /arcanea in Claude Code, the skill activates automatically. The world-state JSON is read. The Eldrian definitions are loaded. The visual DNA is injected. I am building in a world that already exists — not starting from scratch every session.
This is what ACOS enables at a systems level: not just assistance with tasks, but infrastructure that makes every session more coherent than the last. The intelligence compounds. The world deepens.
The broader context for how this fits into my overall AI architecture is in the FrankX Intelligence Ecosystem guide — Arcanea is one output node in a much larger system.
Every Arcanea image follows the same generation protocol:
assets/profile-banner-v2.png for tonal consistencyThe visual DNA specification lives in data/brand-visual-dna.json alongside the main website's visual standards. Same system. Same file. World-building visuals are treated with the same rigor as product visuals.
The result: five character images that read as part of the same world even though they were generated in different sessions. Aethelin's liquid starlight reads as the inverse of Korghast's obsidian alloy. Zyranthis's anti-matter void is visually consistent with Velmara's quantum glass — both at the transparent end of the material spectrum, but for different reasons.
Consistency at this level requires systems, not talent.
Arcanea has a dedicated n8n workflow — "Arcanean X Story Engine" (ID: syLlGYHRRJSu2oqs) — designed to generate and publish lore content automatically.
The pipeline:
data/arcanea-content-queue.json (6 seeded lore episodes, draft status)XAI_API_KEY) for 4-tweet thread generationThe current pipeline has human review built in. This is intentional. Arcanea's lore is canonical — once something is posted publicly, it becomes part of the world-state. A bad AI-generated tweet that contradicts established facts would require a retcon, which is far more expensive than a 30-second approval step.
For GenCreator-level content operations, this is the pattern: automate generation, automate formatting, keep humans in the loop for anything that touches public record. The automation saves the work. The human step preserves the integrity.
The failure mode of AI-assisted world-building is not bad output. The failure mode is inconsistency over time — a world that contradicts itself, characters whose personalities drift, visual styles that fragment across sessions.
The solution is the same as the solution for any distributed system: shared state, deterministic access patterns, write-once canonical definitions, and validation before publication.
Arcanea treats every creative decision as a data write. Every session reads state before producing output. Every new canonical fact goes back into the world-state JSON. The n8n pipeline validates against established lore before anything goes public.
This is not how most people think about creative work. But it is how you build something that scales — a world that can support years of content, multiple characters, multiple story threads, and multiple AI models without collapsing into contradiction.
The research on AI creative tools documents the tools and patterns behind this approach in more detail. The core insight transfers beyond Arcanea: any long-form creative project benefits from treating its lore as a database, its characters as schemas, and its visual identity as a style guide.
Build the infrastructure first. The stories follow naturally.
The core stack is Claude Code (for orchestration and writing), Gemini 3 Pro Image (for character art at 4K), n8n (for automated publishing pipelines), and standard JSON files for world-state persistence. No specialized software required — the architecture is the differentiator, not the tools.
Every ACOS skill session that touches Arcanean content reads from data/arcanea-world-state.json before producing any output. Canonical facts — Eldrian definitions, Godbeast mappings, the nature of the Schism — are injected into context automatically. New content can only extend what is already established, not contradict it, because the established facts are always visible.
Yes. The Eldrian conclave image (public/images/arcanea/eldrian-conclave-20260301.png) establishes their visual relationships when grouped. The world-state JSON documents that all five are distinct from each other in material essence — so a story that involves multiple Eldrians can contrast their natures intentionally rather than arbitrarily.
The Schism is not a war between factions — it is a complementary cycle built into Arcanean cosmology. Understanding this distinction changes how stories within the world work: characters do not choose sides in the Schism; they find their position within an ongoing process. This was established as a canonical fact to prevent the world from defaulting to a predictable light-vs-dark binary.
Arcanea is a creative application of ACOS principles. The same infrastructure that makes AI work sessions coherent and self-improving also makes world-building sessions coherent. The skill layer, the state persistence, the automated pipelines — all of it comes from the same architectural thinking. Arcanea is what happens when you apply enterprise AI framework principles to a fantasy civilization.
Step-by-step guide to setting up ACOS, creating your first agent, and shipping real products with AI.
Start buildingDownload AI architecture templates, multi-agent blueprints, and prompt engineering patterns.
Browse templatesConnect with creators and architects shipping AI products. Weekly office hours, shared resources, direct access.
Join the circleRead on FrankX.AI — AI Architecture, Music & Creator Intelligence
Weekly field notes on AI systems, production patterns, and builder strategy.

The system design decisions that matter when one person builds what teams used to require — from monolith to agents.
Read article
Most skill libraries are noise. These 10 skills changed how I ship code, content, and products — with install commands and real examples.
Read article
The file Claude Code reads first every session. Structure, anti-patterns, and the exact template I use to ship 170+ pages.
Read article