DyCraft：面向长时程 4D 场景生成的智能体系统
原标题：DyCraft: An Agentic System for Long-Horizon 4D Scene Generation
AI 摘要
斯坦福等机构的研究者提出 DyCraft，一个面向长时程 4D 场景生成的智能体系统。它借鉴传统动画的姿势到姿势（pose-to-pose）工作流，把动画生成拆解为规划、3D 分镜关键帧创作与逐帧物理验证三个阶段，让编码智能体在已验证的静态 3D 状态之间生成局部运动。在 24 个长时程多镜头动画任务上，相比 Codex 中的原生 GPT-6 智能体及已有智能体基线，DyCraft 显著减少物体穿插、缺乏支撑等物理违规，生成质量更受人类评审偏好。
正文节选
DyCraft: An Agentic System for Long-Horizon 4D Scene GenerationThanks: Co-first author. Work done when H. Dai was a visiting student at Stanford. †Equal advising. Correspondence to jefftan1@stanford.edu. Project page: https://dy-craft.github.io Abstract While frontier coding agents can generate high-quality 3D scenes and articulated assets, they struggle to create coherent long-horizon animations. When equipped with general-purpose harnesses, they frequently produce physically implausible output