内容来源：https://www.wired.com/story/tecent-3d-models-video-game-design-artificial-intelligence/

内容总结：

近日，腾讯旗下热门团队射击游戏《无畏契约》成为人工智能前沿技术的重要试验场。据知情研究人员透露，游戏开发商拳头公司正利用其母公司腾讯研发的3D原生AI模型“混元”，快速生成游戏角色、场景及剧情原型。

与仅能生成文本、图像的AI模型不同，腾讯“混元”系列模型能直接创造3D物体与交互式场景。除《无畏契约》外，该技术已应用于腾讯旗下另一款游戏《GKART》及部分独立开发团队。腾讯方面对此未予置评。

“游戏行业通常需要大量投入，”该知情人士表示，“过去设计一个角色可能需要一个月，现在只需输入文字描述，‘混元’就能在60秒内提供四个方案。”这一进展标志着，能够理解并重构物理世界的AI模型或将成为游戏设计的标配工具。此类技术不仅将革新游戏内容创作，也为虚拟现实、增强现实及机器人训练开辟了新路径。

普林斯顿大学研究生亚历山大·雷斯特里克指出：“当前3D视觉研究正呈现爆发式增长。其关键应用领域广泛，涵盖内容创作、自动驾驶以及增强现实所涉及的一系列复杂问题。”他特别强调，游戏开发是3D AI模型的天然应用场景，“生成3D网格（表示三维物体的标准方式）正是游戏开发的基础环节。”

然而，AI在游戏创作中的应用亦引发争议。行业普遍担忧AI可能导致岗位缩减，部分开发者主张对AI生成内容进行标识，亦有观点认为该技术已在业内普及，规范为时已晚。

今年7月，腾讯发布可生成交互场景的“混元世界1.0”模型，随后在10月推出支持视频转3D场景的新版本。测试显示，该模型能快速构建色彩明丽的虚拟场景及可3D打印的定制化角色。

腾讯的探索折射出AI研究的整体转向：众多专家认为，AI需深化对物理世界的理解才能实现突破。目前微软、Meta、Stability AI及字节跳动等企业均布局3D AI模型，而腾讯“混元”在相关评测中表现领先。初创公司亦在该领域积极创新，如斯坦福大学李飞飞教授创立的World Labs开发的“Marble”工具，能生成高度一致的持久化3D场景，为即时游戏生成及机器人训练提供支持。

学术界同样活跃：斯坦福大学的“3D通才”项目利用大语言模型决策场景编辑，普林斯顿大学研究者正探索通过代码生成3D场景，谷歌DeepMind的SIMA 2项目则展示了AI智能体与虚拟世界交互创造新玩法的前景。

随着3D AI重要性日益凸显，腾讯凭借其在全球游戏、影视领域的深厚积累，以及微信生态与自研AI助手“元宝”的协同优势，有望在竞争激烈的中国AI赛道中占据独特地位。尤其在迈向3D化的AI时代，其游戏开发经验或将成为关键竞争力。

中文翻译：

快节奏团队射击游戏《无畏契约》近期已成为人工智能研究一个前景广阔新方向的试验场。据一位熟悉相关工作的匿名研究人员透露，该游戏的开发商拳头游戏（腾讯子公司）正在运用原生3D人工智能模型进行新角色、场景与剧情线的原型设计。

尽管许多人工智能模型已能生成文本、图像与视频，但腾讯的混元模型家族已能构想出3D物体与交互式场景。消息人士称，腾讯旗下另一款游戏《GKART》的开发团队以及部分独立开发者也在使用这些模型。腾讯对此不予置评。

"游戏行业需要大量投入，"该消息人士表示，"过去设计一个角色需要一个月，现在只需输入文字描述，混元模型就能在60秒内提供四种方案。"

这则消息释放出一个早期信号：能够理解并重构物理世界的人工智能模型可能成为游戏设计的标准配置。除了生成游戏内容，这类模型还能推动虚拟现实与增强现实技术发展，并帮助机器人学习新技能。

"当前3D视觉研究呈现爆发式增长，"普林斯顿大学研究生亚历山大·雷斯特里克表示，他正致力于开发生成3D内容的新方法，"杀手级应用层出不穷：内容创作、自动驾驶，以及增强现实涉及的整条技术链。"

雷斯特里克补充说，电子游戏是3D人工智能模型的天然应用场景。"输出3D网格（呈现3D物体的标准方式）本就是游戏开发的基础环节。"

但与其他创意领域类似，运用人工智能开发游戏存在争议。对人工智能导致失业的忧虑日益凸显。部分开发者主张含有AI生成内容的游戏应进行标注，另一些人则认为为时已晚——这项技术已在行业内无处不在。

腾讯于今年7月发布了可生成交互场景的混元世界1.0模型。数月前笔者曾测试该模型，探索过一个宛如乐高电影场景的彩色积木山谷。近期则体验了更基础的混元3D模型，它能幻化出各类3D物体。笔者用它生成了一些精美的《龙与地下城》定制角色用于3D打印。10月，腾讯发布了混元世界新版，支持用户上传视频生成3D场景。

腾讯混元模型标志着人工智能研究领域正在发生更广泛的转变。许多专家认为，人工智能模型需要深化对物理世界的理解才能取得突破。正因如此，腾讯并非唯一研发原生3D人工智能模型的企业。微软、Meta、Stability AI和字节跳动都推出了3D模型，但混元在相关评测榜单中位居榜首。

多家初创公司也在该领域展开有趣探索。由现代人工智能建设关键人物、斯坦福计算机科学家李飞飞创立的World Labs公司，开发出名为Marble的工具，能生成完全一致且持久存在的3D场景。这项技术既可用于实时游戏生成，也能为机器人提供可靠的训练数据。

3D人工智能同样是学术研究的热点领域。斯坦福大学的"3D通才"项目利用大语言模型决策如何用新物体改造场景；普林斯顿研究生雷斯特里克正在开发通过代码生成3D场景的技术，使大语言模型能以更强大的方式生成场景并与之互动；而谷歌DeepMind的SIMA 2等项目则展示了智能体如何通过与虚拟世界交互创造新型游戏玩法。

随着3D人工智能的重要性日益凸显，在众多竞逐该领域的中国人工智能企业中，腾讯可能成为日益重要的参与者。除了出品全球热门电子游戏与电影，腾讯还运营着在中国无处不在的超级应用微信，并推出了集成于微信的智能助手元宝。在这个日益立体化的人工智能世界里，腾讯的游戏开发专长或许将成为其独特优势。

本文节选自威尔·奈特《人工智能实验室》通讯，过往内容可通过此处查阅。

英文来源：

The video game Valorant, a fast-paced team-based shooter, has recently become a testing ground for a promising new direction in artificial intelligence research. The game’s developers at Riot Games (a Tencent subsidiary) are using 3D-native AI models to prototype new characters, scenes, and storylines, according to a researcher familiar with the company’s efforts who spoke on the condition of anonymity.
While many AI models can generate text, images, and video, Tencent’s Hunyuan (混元 or “first mix”) family of models can dream up 3D objects and interactive scenes. The source says that Tencent’s models are also being used by the developers of another Tencent game, GKART, and by some independent developers, too. Tencent declined to comment.
“The games industry requires a lot of investment,” the source says. “Previously you would need a month to design a character. Now you can just type in some text, and Hunyuan can give you four choices in 60 seconds.”
The news is an early signal that models capable of understanding and re-creating the physical world could become a standard ingredient in game design. In addition to generating game content, these models could also enable more advanced virtual and augmented reality and help robots learn to do new things.
“There’s a real explosion of 3D vision research nowadays,” says Alexander Raistrick, a graduate student at Princeton University working on novel approaches to generating 3D content. “There are many killer applications: There's content creation, there’s self driving, and there’s a whole stack of problems involved in augmented reality.”
Raistrick adds that video games are an obvious application for 3D AI models. “Outputting 3D meshes [a standard way of representing 3D objects] is your typical kind of bread and butter of game development,” he says.
But, as in other creative fields, using AI to create video games is controversial. Concerns about AI-fueled job loss loom large. Some developers say games should be labeled when they contain AI-made content. Others say it’s too late: The technology is already ubiquitous in the industry.
Tencent released HunyuanWorld 1.0, a model that generates interactive scenes, in July. I tested it a few months back, exploring a scene that looked like it was part of a Lego movie—a valley of brightly colored blocks disappearing into the distance. More recently, I’ve been playing with a more basic model, Hunyuan 3D, which can conjure up 3D objects. I used it to generate some very nice custom Dungeons & Dragons characters to 3D print. In October, Tencent released a new version of HunyuanWorld that lets users upload video to generate 3D scenes.
Tencent’s Hunyuan models point to a broader shift happening in AI research. Many experts believe that AI models will need a deeper understanding of the physical world to advance. Because of this, Tencent is far from alone in building 3D-native AI models. Microsoft, Meta, Stability AI, and Bytedance all offer 3D models, but Hunyuan sits at the top of one leaderboard designed to rank such tools.
A number of startups are doing interesting work in this space, too. World Labs, founded by Fei-Fei Li, a Stanford computer scientist who played a key role in building modern AI, has developed a tool called Marble that produces fully consistent and persistent 3D scenes. This could be useful for generating games on the fly or producing reliable training data for robots.
3D AI is also an exciting area for academic research. A Stanford University project called 3D Generalist used an LLM to decide how to modify scenes with new objects. Raistrick, the graduate student at Princeton, is developing a way of generating 3D scenes using code, an approach that makes it possible for LLMs to generate and interact with scenes in a more powerful way. And projects like Google DeepMind’s SIMA 2 show how AI agents could interact with virtual worlds to create new forms of gameplay.
As 3D-capable AI becomes more important, Tencent may emerge as an increasingly important player among a host of Chinese AI firms clamoring to win in this space. Besides producing some of the world’s most popular video games and movies, it operates WeChat, a chat app with a wide range of other functions, that is ubiquitous in China. Tencent also has its own chatbot, called YuanBao, which is integrated into WeChat. But Tencent’s video game skills may give it a distinct edge in an increasingly 3D AI world.
This is an edition of Will Knight’s AI Lab newsletter. Read previous newsletters here.