If Vibe Coding made this thing about "writing code" from "touching the keyboard" to "talking to AI", then the work we're going to present today may be repeating the same story in the field of building a 3D world.
A team from the Hong Kong University of Technology (HQ) Joint Tencent AI Platform has proposed VibeWorlding: a way to write code like an Agent, through multiple rounds of dialogue, calling tools, observing rendered feedback from a multi-modal 3D world framework that builds and edits interactive 3D. All you have to do is say, "Give me a messy desert temple", or "Put the chair next to the table two meters forward", and the rest is to retrieve 3D assets, place objects, avoid collisions, render confirmation, and hand it over to Agent.
The paper is titled: VibeWorlding: Can Multimodal Agents Construct 3D Open Worlds.