Research map / 3D vision / 3D generation
SynCity: Training-Free Generation of 3D Worlds
Paul Engstler, Aleksandar Shtedritski, Iro Laina, Christian Rupprecht, Andrea Vedaldi
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add rebasing (larger gray base under tile image) before image-to-3D (ablation) | SynCity · Tile geometry (2x2 scene) | squareness: 0.92 -> 1.00 | Improved |
| add rebasing (larger gray base under tile image) before image-to-3D (ablation) | SynCity · Tile geometry (2x2 scene) | completeness: 0.73 -> 1.00 | Improved |
| add rebasing (larger gray base under tile image) before image-to-3D (ablation) | SynCity · Tile geometry (2x2 scene) | base area (voxels): 2271 -> 4096 | Improved |
| use proposed latent upsampling with a single conditioning frame instead of naive interpolation (ablation) | SynCity latent upsampling · Latent upsampling (200 tiles) | FID: 200.5 -> 111.6 | Improved |
| use multiple conditioning views instead of a single frame in latent upsampling (ablation) | SynCity latent upsampling · Latent upsampling (200 tiles) | FID: 111.6 -> 89.1 | Improved |
| use multiple conditioning views instead of a single frame in latent upsampling (ablation) | SynCity latent upsampling · Latent upsampling (200 tiles) | LPIPS: 0.3517 -> 0.3212 | Improved |
| use proposed multi-frame latent upsampling instead of naive interpolation (ablation) | SynCity latent upsampling · Latent upsampling (200 tiles) | KID: 0.243 -> 0.051 | Improved |
| use proposed multi-frame latent upsampling instead of naive interpolation (ablation) | SynCity latent upsampling · Latent upsampling (200 tiles) | SSIM: 0.3093 -> 0.5312 | Improved |
| add context from neighbouring tiles to 2D prompting (ablation) | SynCity · 2x2 grid (qualitative) | scale consistency of buildings | Improved |
| add 3D blending of tile boundaries (ablation) | SynCity · generated scene (qualitative) | tile continuity | Improved |
| add isometric base image and mask framing to 2D prompting (ablation) | SynCity · tile generation (qualitative) | tile framing suitability | Improved |
| generate whole scene non-iteratively in one image instead of tile grid (ablation) | Flux + TRELLIS · city scene (qualitative) | layout control and detail | Hurt |
Later papers that compared with this method
4 later accepted papers report SynCity as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.
- WorldGrow: Generating Infinite 3D World (AAAI 2026 · better on 6 of 6 shared results)
- Map2World: Segment Map Conditioned Text to 3D World Generation (ECCV 2026 · better on 5 of 6 shared results)
- Extend3D: Town-Scale 3D Generation (CVPR 2026 · better on 4 of 4 shared results)
- Yo'City: Personalized and Boundless 3D Realistic City Scene Generation via Self-Critic Expansion (CVPR 2026 · better on 3 of 3 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- You See it, You Got it: Learning 3D Creation on Pose-Free Videos at Scale (CVPR 2025 · similarity 0.83)
- WonderWorld: Interactive 3D Scene Generation from a Single Image (CVPR 2025 · similarity 0.82)
- WorldCoder-Bench: Benchmarking Physically Grounded 3D World Synthesis (NeurIPS 2026 · similarity 0.82)
- WorldAct: Activating Monolithic 3D Worlds into Interactive-Ready Object-Centric Scenes (NeurIPS 2026 · similarity 0.82)
- MVPaint: Synchronized Multi-View Diffusion for Painting Anything 3D (CVPR 2025 · similarity 0.81)
- 4DWorldBench: A Comprehensive Evaluation Framework for 3D/4D World Generation Models (CVPR 2026 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
