Sakana AI and NYU Release Dream-Cubed: AI Generates Playable Minecraft Worlds Trained on Billions of Cubes
tkasasagi · x · 2026-07-30
Sakana AI, in collaboration with New York University, introduces Dream-Cubed, a large-scale dataset and family of generative models for generating Minecraft worlds at block resolution. The dataset comprises billions of high-quality cubes from procedurally generated terrain and human-authored maps. They trained discrete and continuous 3D diffusion models for biome-conditioned chunk generation. The discrete masked diffusion formulation enables inpainting, outpainting, and user-defined block conditioned generation, allowing players to edit and play generated structures immediately. Paper, code, and blog are publicly available.
More from Multimodal
- Tessellation-free geometric queries on displaced surfaces cut caustics variance 10x — ssh4net · 2026-10-16
- Complex multi-link CAD mechanism finally animates, dev celebrates the breakthrough — zetalyrae · 2026-10-16
- OpenMOSS releases MOSS-Image-SFT: Apache 2.0 text-to-image model on Hugging Face — bdsqlsz · 2026-10-16
- Hyperflow open-sourced: 49 sampling steps down to 8, ~3x faster video generation — mhdfaran · 2026-10-16
- Cinematic 3D fantasy character prompt template with salvaged-robot armor — azed_ai · 2026-10-16
- Midjourney vintage woodcut style: --sref 2446102396 beaver with a cleaver — michaelrabone · 2026-10-16