Sakana AI Introduces Dream-Cubed: 3D Generative Models for Minecraft Trained on Billions of Blocks
hardmaru · x · 2026-07-30
Sakana AI has introduced Dream-Cubed, a new large-scale dataset and a family of generative models designed to generate high-fidelity Minecraft worlds at block resolution.
- Data & Models: The researchers compiled billions of carefully balanced voxels from procedurally generated and human-authored maps to train both discrete and continuous 3D diffusion models.
- Controllable Generation: The discrete masked diffusion formulation not only handles biome-conditioned chunk generation but also natively supports inpainting, outpainting, and user-defined block conditioning.
- Practical Application: This enables players and creators to instantly generate editable and playable structures and terrains, bridging the gap in generative AI for interactive 3D worlds.
More from Multimodal
- Seedance 2.5 workflow: character reference sheets + structured prompt yields a 30s single-take fake tourist video — techhalla · 2026-10-16
- AI couple moves into a castle ruin: full Magnific + Seedance 2.5 prompt thread behind the viral clip — techhalla · 2026-10-16
- FLUX.3 Klein shines at consistent high-res photo editing on RTX 5090: 3000×4000 in ~230s — herbertseabra · 2026-10-16
- Smoky Velvet Noir: a reusable image-gen prompt template with three moody examples — LudovicCreator · 2026-10-16
- VideoGen-Agent: An 8B Multimodal Agent Turns Video Generation Into a Tool-Use Loop — jiqizhixin · 2026-10-16
- Redditor builds impressive seaplane sim project with Sonnet 5.5 + Blender — Wise_Shame_2853 · 2026-10-16