Programmable World Model paper gives AI-generated games persistent, rule-based world state
mhdfaran · x · 2026-09-10
- A new paper from Alaya Lab, Programmable World Model (arXiv:2609.10540), tackles the core flaw of AI-generated games: video world models produce believable next frames but can't remember persistent facts — which NPCs are alive, HP counts, what you're carrying.
- The key idea is decoupling world-state evolution from visual generation: an agent translates natural-language instructions into executable programs defining entity states and transition rules; a lightweight engine runs them to maintain an explicit, persistent global world state, including off-screen entities and non-visual attributes.
- State-augmented 3D oriented bounding boxes (OBBs) serve as the intermediate representation, deterministically compiled with the camera trajectory into pixel-aligned spatiotemporal conditioning for a pretrained video model acting as generative renderer — enabling playable games with predefined mechanics and direct entity control.
- The authors also release CombatStateBench, where the method hits 94% Count Accuracy and 98% State Accuracy, substantially outperforming existing interactive approaches.
Related event: Alaya Lab's Programmable World Model Gives AI Games Persistent State(2 posts)→
More from Multimodal
- Dataset Distillation Method Turns Artist Catalogs into Stunning Composites at ECCV Art Gallery — vincesitzmann · 2026-09-10
- ByteDance's Seedance 2.0 Video Model Now Available on fal with US-Only Hosting — isidentical · 2026-09-10
- Jerry Springer x Sailor Moon AI parody: creator shares MiniMax, Kinovi, Fish.audio workflow — Ok-Giraffe-8670 · 2026-09-10
- Gaussian Wrapping: image-to-3D mesh reconstruction presented at ECCV 2026 poster 310 — ducha_aiki · 2026-09-10
- Blogger tests DeepSeek on 3D model generation, says the results speak for themselves — kevinkern · 2026-09-10
- Illustration web-app built on tldraw: sketch, pick a style, generate code, refine — threepointone · 2026-09-10