DyCraft: An Agentic System for Long-Horizon 4D Scene Generation

Computer Science > Computer Vision and Pattern Recognition arXiv:2610.40754 (cs) [Submitted on 29 Oct 2026] Title: DyCraft: An Agentic System for Long-Horizon 4D Scene Generation Authors: Jeff Tan, Hang Dai, Zeqian Long, Yanzhe Lyu, R. Kenny Jones, Shangzhe Wu, Chen Geng, Jiajun Wu.

计算机科学 > 计算机视觉与模式识别 arXiv:2610.40754 (cs) [提交于 2026 年 10 月 29 日] 标题：DyCraft：用于长时程 4D 场景生成的代理系统 作者：Jeff Tan, Hang Dai, Zeqian Long, Yanzhe Lyu, R. Kenny Jones, Shangzhe Wu, Chen Geng, Jiajun Wu。

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 outputs, generating objects that interpenetrate, lack physical support, or move unnaturally.

摘要：尽管前沿编码代理能够生成高质量的 3D 场景和关节资产，但它们在创建连贯的长时程动画方面仍面临困难。当配备通用工具时，它们经常产生物理上不可信的输出，例如生成相互穿透、缺乏物理支撑或运动不自然的物体。

Such bare agents struggle to repair these errors because they default to scripting the full animation in one pass, making it difficult to verify their work and disentangle complex spatiotemporal constraints. Drawing on the practice of human animators, we introduce DyCraft, a system for animated 3D scene generation that breaks animation into steps that coding agents can more reliably solve.

这些基础代理难以修复此类错误，因为它们默认一次性编写整个动画脚本，这使得验证其工作并解开复杂的时空约束变得困难。借鉴人类动画师的实践，我们引入了 DyCraft，这是一个用于动画 3D 场景生成的系统，它将动画分解为编码代理可以更可靠地解决的步骤。

DyCraft first expands the user’s prompt into a detailed plan and builds a static scene. It then creates a 3D storyboard of key animation poses, where each frame is a static 3D scene that the agent can build and debug in isolation. Storyboarding decomposes the problem into pairs of verified start and end states, allowing the agent to author local motion between them with a collection of motion authoring tools.

DyCraft 首先将用户的提示扩展为详细的计划并构建静态场景。然后，它会创建关键动画姿势的 3D 分镜，其中每一帧都是一个静态 3D 场景，代理可以对其进行独立构建和调试。分镜将问题分解为成对的已验证起始状态和结束状态，允许代理使用一系列运动创作工具在它们之间创作局部运动。

We evaluate DyCraft on 24 long-horizon, multi-shot animation tasks covering humanoid robots, human and animal characters, and physical mechanisms. Compared with a native GPT-6 agent in Codex and prior agentic baselines, DyCraft substantially reduces physical violations and generates higher-quality animated 3D scenes that are more often preferred by human raters.

我们在涵盖人形机器人、人类和动物角色以及物理机制的 24 项长时程、多镜头动画任务上评估了 DyCraft。与 Codex 中的原生 GPT-6 代理及先前的代理基线相比，DyCraft 大幅减少了物理违规现象，并生成了质量更高、更受人类评估者青睐的动画 3D 场景。