---
格式版本: 2
标题: "H3-World: Turning Language Understanding into World Control"
原文链接: "https://arxiv.org/abs/2609.01560"
发布日期: "2026-09-01"
发布时间校准状态: "found"
发布时间需复核: "否"
发布时间来源: "rule:local:strict_original_body"
发布时间证据: "**\\[v1\\]** Tue, 1 Sep 2026 17:21:40 UTC (46,598 KB)"
发布时间校准原因: "规则确认唯一严格发布时间，来源 local:strict_original_body"
发布时间校准置信度: "high"
发布时间候选数量: 18
发布时间严格候选数量: 6
发布时间原页读取状态: "source template page reused from URL open"
发布时间未找到原因: ""
发布时间校准时间: "2026-09-04T01:42:26+08:00"
发布时间仲裁状态: "skipped"
发布时间仲裁尝试次数: 0
发布时间仲裁耗时毫秒: 0
发现时间: "2026-09-04T01:41:38+08:00"
入库时间: "2026-09-03T17:42:27.029Z"
来源平台: "arXiv 学术论文搜索"
搜索渠道: "source_template"
搜索词: "site:arxiv.org 腾讯"
匹配关键词:
  - "AI"
相关厂家:
  - "腾讯"
相关专家:
  []
内容类型: "网页"
抓取工具: "Free Fetch + Defuddle"
清洗工具: "Defuddle Markdown + Defuddle/Readability 正文提取"
原始附件:
  []
AI优质: "否"
AI打分: 10
AI分档: "非优质"
AI质检状态: "不通过"
AI打分理由: "论文主题为视频生成器转交互世界模型，与超节点/AI Rack/机柜级AI基础设施及供电散热互连等主题无关，属明显无关内容。"
AI质检模型: "zj-deepseek-v4-flash"
AI质检时间: "2026-09-04T01:42:34+08:00"
AI主题相关性: 0
AI来源权威性: 5
AI新颖性: 0
AI技术细节: 0
AI商业部署信号: 0
AI完整性: 5
AI摘要: "H3-World 提出一个高效框架，将 33B 的 MiniMax-H3 视频生成器改造为交互式世界模型，通过结构化字符与相机指令及时间注意力路由，实现对角色和镜头的精确、时序化控制。"
AI摘要模型: "ali-deepseek-v4-flash"
AI摘要时间: "2026-09-04T00:17:19.411Z"
采集批次: "2026年9月3日22点43分34秒"
采集批次ID: "20260903-224334-406"
去重键: "https://arxiv.org/abs/2609.01560"
---

## Computer Science > Computer Vision and Pattern Recognition

## Title:H3-World: Turning Language Understanding into World Control

Authors:[Danze Chen](https://arxiv.org/search/cs?searchtype=author&query=Chen,+D), [Zeqing Wang](https://arxiv.org/search/cs?searchtype=author&query=Wang,+Z), [Ziyue Lin](https://arxiv.org/search/cs?searchtype=author&query=Lin,+Z), [Xingyi Yang](https://arxiv.org/search/cs?searchtype=author&query=Yang,+X), [Yeying Jin](https://arxiv.org/search/cs?searchtype=author&query=Jin,+Y)

[View PDF](https://arxiv.org/pdf/2609.01560) [HTML (experimental)](https://arxiv.org/html/2609.01560v1)

> Abstract:We present H3-World, an efficient framework that turns the 33B MiniMax-H3 video generator into an interactive world model. Our key finding is that, as large video generators become more capable, language is emerging as a natural interface for control. MiniMax-H3, for example, already supports zero-shot control of character behavior and camera motion through natural-language instructions. Building on this, H3-World turns this coarse language interface into precise, temporally grounded world control, without introducing dedicated action modules. Specifically, we represent each action as a structured combination of character and camera instructions, and align them with the corresponding temporal video latents. To make the control temporally precise, we further introduce temporal attention routing, which restricts each instruction to its intended time interval and reduces control leakage across actions. Importantly, H3-World directly reuses the semantic representations learned during large-scale video pretraining and requires only lightweight adaptation. With only 8,000 gameplay samples, 10,000 LoRA optimization steps, and 0.199% trainable parameters, H3-World achieves effective character and camera control while preserving strong generation quality. It also generalizes to unseen scenarios. These results show that the control capabilities emerging in large video generators can be efficiently transformed into interactive world control.

| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI) |
| --- | --- |
| Cite as: | [arXiv:2609.01560](https://arxiv.org/abs/2609.01560) \[cs.CV\] |
|  | (or [arXiv:2609.01560v1](https://arxiv.org/abs/2609.01560v1) \[cs.CV\] for this version) |
|  | [https://doi.org/10.48550/arXiv.2609.01560](https://doi.org/10.48550/arXiv.2609.01560) |

## Submission history

From: Danze Chen \[[view email](https://arxiv.org/show-email/41af0934/2609.01560)\]  
**\[v1\]** Tue, 1 Sep 2026 17:21:40 UTC (46,598 KB)

[Which authors of this paper are endorsers?](https://arxiv.org/auth/show-endorsers/2609.01560) | Disable MathJax ([What is MathJax?](https://info.arxiv.org/help/mathjax.html))