📢WorldAgents: 3D worlds only from 2D image models - without any training!
Matthias Niessner
Visit on X@MattNiessner50K followers
19,093 views
📢WorldAgents: 3D worlds only from 2D image models - without any training!
We propose an agentic approach with a Director (VLM) to plan the scene, a Generator (Flux or NanoBanana) for new views, and a Verifier (VLM) for selection / 3D consistency.
-> High-fidelity 3D worlds from a single text prompt.
What's remarkable: our agents find consistent views from 2D image models to obtain 3D-consistent worlds; this shows that image models contain world priors - agents just need to find them!
Great work by Ziya Erkoç Angela Dai
4 Comments
Comments from the original post on X (Twitter)

@K_S_Schwarz
@ErkocZiya Our work [2503.16611] A Recipe for Generating 3D Worlds From a Single Image was among the first to show how to use image models like Flux to generate consistent worlds. You are missing the citation in the related work and attribution of credit.
1 like

@LionTV_crypto
agents cooking 3D from 2D this is insane alpha drop
1 like

@Elon_Orbit
Director-Generator-Verifier pipeline. Solid. But anyone seen the compute bill? That's where these papers get quiet.

@Jenny3Brauner
nice
