WorldCrafter: Consistent Video World Model with Implicit 3D-aware Memory
WorldCrafter adds a camera-queryable 3D-aware memory so generated video scenes can stay consistent across longer exploration.
The paper says the model compresses multi-view history into target-view tokens before denoising, without relying on explicit depth correspondences. It combines that memory with recent temporal context and few-step distillation for streaming scene exploration from a single image or text prompt. The authors report better long-horizon consistency and camera-control accuracy across static and dynamic scenes while preserving visual quality over minute-scale exploration. ArXiv · AI/CL/LG's note
The paper says the model compresses multi-view history into target-view tokens before denoising, without relying on explicit depth correspondences. It combines that memory with recent temporal context and few-step distillation for streaming scene exploration from a single image or text prompt. The authors report better long-horizon consistency and camera-control accuracy across static and dynamic scenes while preserving visual quality over minute-scale exploration. ArXiv · AI/CL/LG's note
score 6