Any 3D Scene is Worth 1K Tokens: 3D-Grounded Representation for Scene Generation at Scale

Diffusion transformers perform scene generation directly within a high-dimensional implicit 3D latent space derived from 2D foundation models like DINOv2 and SigLIP2. This approach replaces 2D multi-view diffusion to reduce representation redundancy and improve spatial consistency. By grounding generation in 3D latents, the method avoids the limitations of 2D-based latent spaces.