SynCity 3000: Bootstrapping Scene-Scale 3D Diffusion

Forum topic · 小凯 · 2026-07-08

Summary

SynCity 3000 is a 3D scene generation framework that produces globally consistent, layout-controllable scenes at scale. Building on modern image-to-3D generators that can create complex 3D assets from a single image, the authors adapt such a generator into a convolutional operator, extending its capability from single assets to entire scenes. To address the scarcity of 3D scene training data, the model is fine-tuned on scene-level data produced by a new synthetic data engine. The convolutional generator is applied to an axonometric (isometric-style) image of a full scene generated from a user prompt, yielding 3D scenes of arbitrary scale and complexity. Across diverse prompts and layouts, SynCity 3000 generates large, coherent, and detailed scenes, overcoming limitations of prior 3D scene generation methods. The paper is authored by Paul Engstler, Iro Laina, Christian Rupprecht, and Andrea Vedaldi, and is available as arXiv:2607.05392.

Paper Overview

Field: Computer Vision (CV) Authors: Paul Engstler, Iro Laina, Christian Rupprecht, Andrea Vedaldi Published: 2026-07-06 arXiv: 2607.05392

Abstract

This paper presents SynCity 3000, a framework for 3D scene generation that is both globally consistent and amenable to fine-grained layout control.

Key Ideas

Leverage existing image-to-3D generators: Current generators can already produce complex 3D assets from a single image. SynCity 3000 builds directly on this capability.
Convolutional adaptation: The image-to-3D generator is adapted into a *convolutional operator*, extending asset-level generation to scene-scale generation.
Synthetic data engine: Since large-scale 3D scene training data is scarce, the authors introduce a new synthetic data engine that generates scene-level training data, on which the model is fine-tuned.
Axonometric rendering: The convolutional generator is applied to an axonometric image of an entire scene synthesized from a user prompt, producing 3D scenes of arbitrary scale and complexity.

Results

Across diverse prompts and layouts, SynCity 3000 generates large, coherent, and detailed 3D scenes, addressing shortcomings of previous 3D scene generation methods.

--- *Auto-collected on 2026-07-06*