WM^MLLM: Princeton and NVIDIA turn multimodal LLMs into editable world models
OpenEnvisionLab · hf · 2026-11-20
WM^MLLM from Princeton, NVIDIA, and OpenEnvision is the first interactive world model built on multimodal LLMs, generating persistent, editable worlds from a single image.
- Key design: keep the world state outside the generator as a panorama with revision history, written by the MLLM — it turns each request into operations on the state, selects where local edits are written, and checks every candidate before it becomes the next state
- Unlike video world models where revisits are resynthesized and edits drift, image generators only propose views and video only presents them; WM^MLLM guarantees exact revisits, lets users walk to new places and return, edit objects via text or a box, place subjects, and change lighting, weather, or season
More from Multimodal
- Coding agent demo: 16 minutes for a Three.js macro scene, 92 agents for a 3D world — Saboo_Shubham_ · 2026-11-20
- AI turns Mediterranean illustration into interactive 3D world; Three.js wall-crash demo costs $6.98 — Saboo_Shubham_ · 2026-11-20
- Kling team's Coda-DaVinci reimagines video generation via Code2Video agents — KlingTeam · 2026-11-20
- Tencent's WeMM-Video pushes open-source omni-modal video understanding — tencent · 2026-11-20
- Kling's EgoCaptioner uses 50k samples and GRPO to master egocentric video captioning — KlingTeam · 2026-11-20
- Open-source repo offers 500+ AI-generated low-poly 3D landmarks for maps under CC0 — tomjohndesign · 2026-11-20