Once an AI walks into a 3D world, it cannot see the scene — it knows the world through structured information the system feeds it. The most important piece is the "AI description" attached to each object.

How well that description is written decides what it can say when standing in front of an exhibit: written clearly, it knows "this is machine No. 3, used for machining this part, watch the guide rails"; written poorly, all it knows is "there is something there." This article explains this entry point thoroughly: how the mechanism works, what the rules are, and how to write a good one.

One: how AI "sees" an object

First, the mechanism: the AI renders nothing and has no camera-style vision. Its knowledge of the world comes from three inputs:

The radar tells it "there is an object there"; the AI description tells it "that object is a …". Only together do they let it say "the machine in front of you" instead of "that thing in front of you".

Two: the rules — one field, one limit, one truncation

Three hard rules, baked into the system:

1. Filled in the editor: every object's property panel has an AI description field — write it while building the scene;
2. Up to 500 characters: the field's limit;
3. Truncated to 300 characters when delivered with the radar: the easily missed one — you may write 500, but the AI only receives the first 300. The most important information must be inside the first 300 characters.

Three: how to write it — a three-part structure

A structure that works in practice:

Part 1 · What it is (first, always within 300 characters)

"A five-axis machining center, displayed here to show the machining process of this part." — one sentence of definition, no throat-clearing.

Part 2 · Where it is, what it sits next to

"East side of the exhibition hall, next to machine No. 4 and the visitor entrance." — the radar already knows relative positions; a short orientation note lets the AI chain multiple objects into a route, which is what makes a guided tour coherent.

Part 3 · What visitors should look at

"Focus on the tool changer and the guide rails; specifications are on the placard nearby." — give it one sentence it can actually say out loud.

Counter-example: a description that reads "model 1", "exhibit", "device" — that is the same as writing nothing. The ceiling of the AI's explanation was set the moment you wrote the description.

A practical ratio: two-thirds of the text on "what it is", one-third on "where and what to look at". Orientation and highlights must be short, but they must be there.

Four: the description is not everything — how the three inputs work together

The object description solves "what is this"; a decent explanation also needs:

• Knowledge base: deeper parameters, fault trees, FAQs — organized locally by the enterprise and retrieved by the AI client at answer time; the platform does not host knowledge bases;
• Chat context: it remembers what a visitor asked before (recoverable on reconnect);
• Event stream: three configurable levels (off / aggregated per second / per-event), deciding how fast it reacts to changes in the scene.

The description is the entry point, not the whole. Write the entry point well, and the depth sitting in your knowledge base finally gets a chance to be spoken.

Five: boundaries, as always

• The AI cannot see: the object description is the primary text channel through which it knows an object; visual effects are invisible to it;
• Truncation at 300 characters: everything from character 301 onward never reaches the AI;
• No hosted knowledge base: the platform provides three inputs (descriptions, chat history, radar) only;
• Templated docents (pick a template in the backend, fill in a knowledge base, get a ready docent) are still on the roadmap, not implemented — today, descriptions are written object by object in the editor.

Six: FAQ

Q: Can I write the description in English?

A: The system is bilingual (Chinese/English). The description is delivered to the AI in the language it was written in. For worlds serving Chinese visitors, Chinese is the safer choice.

Q: How many objects need descriptions?

A: Cover the key stops on the route — not every object. An exhibition route typically has a dozen or so key points; objects without descriptions are known to the AI only as positions, without content.

Q: Can descriptions be edited later?

A: Yes. Edit them in the editor; they save with the scene and the next radar delivery carries the new text.

Q: How is this different from just feeding documents to an LLM?

A: The difference is that it knows where the visitor is and what is in front of them. Object descriptions plus the spatial radar anchor every answer to a specific spot — the same model, fed a different information structure, produces a completely different explanation.

Seven: Source and repositories

The three addresses have identical content; the first two are faster for visitors in mainland China. The repositories include deployment guides and a demo entry.

• Gitee (faster in mainland China): https://gitee.com/miduoxinxijeji/miduo.git
• GitCode (mirror): https://gitcode.com/qq_35054471/virtual-world
• GitHub: https://github.com/miduo100/3d-virtual-world

About Genesis

Genesis is a self-hosted 3D virtual world system built on Three.js + WebGL, helping individuals and businesses build their own 3D spaces. Accessible directly from a browser, compatible with both PC and mobile, it supports multiplayer online, federated teleportation, a shop system, and Agent integration—where an AI can enter your world as an embodied character. Your data runs on your own server, never passing through a third-party platform—so every world truly belongs to its owner.

Want the AI to explain your world clearly? Start by writing clearly the text it will read. Genesis (创世虚拟世界CRM系统) is a Three.js 3D virtual world foundation that deploys on your own servers — every object in the editor carries an AI description, and with the spatial radar, the AI's explanations anchor to the spots you choose. The official site (search for 创世虚拟世界CRM) has a demo world you can walk around in.

About the name: Genesis in this article refers to 创世虚拟世界CRM系统 — the same self-hostable 3D virtual world product. If searching "Genesis" doesn't find us, search for 「创世虚拟世界CRM」.