Bold claim: AI is quietly rewriting how games are designed, and the impact could reach far beyond the screen. A stirring example comes from Valorant, Riot Games’ fast-paced, team-based shooter, where developers are piloting 3D-native AI models to prototype new characters, scenes, and storylines. A source close to the project—speaking on condition of anonymity—says Tencent’s Hunyuan family of models can generate 3D objects and interactive environments in minutes, a stark contrast to the old workflow that often took weeks or longer. Tencent shows the tech in action not only for Valorant but also for GKART and several independent studios, though Tencent has declined to comment formally.
The promise is clear: AI that can understand and recreate the physical world could become a standard tool in game creation. Beyond content generation, these models might accelerate advances in virtual and augmented reality and even teach robots to perform new tasks. Alexander Raistrick, a Princeton graduate student researching 3D content generation, notes that 3D vision research is booming with a wide range of applications—from content creation to self-driving to extensive AR pipelines—and video games are a natural home for 3D AI models. He explains that producing 3D meshes is a core, bread-and-butter activity in game development.
Yet, the excitement is matched by controversy. The rise of AI-driven game design raises concerns about job disruption within the industry. Some developers advocate for clear labeling when content is AI-generated, while others argue that AI integration in game production is already a prevailing reality and may be irreversible.
Tencent released HunyuanWorld 1.0, a model capable of generating interactive scenes, in July. Personal testing revealed a scene reminiscent of a Lego world—bright blocks receding into a distant valley. A later iteration, Hunyuan 3D, focuses on creating 3D objects, and more recently an update enables users to upload video to drive 3D scene generation. These capabilities position Hunyuan as part of a broader trend toward 3D-native AI in AI research.
Major tech players are exploring similar directions. Microsoft, Meta, Stability AI, and Bytedance all offer 3D modeling capabilities, with Hunyuan competing for leadership in this space on leaderboards that track 3D model quality and usefulness. Startups are contributing as well; for example, World Labs, led by Fei-Fei Li of Stanford, has built Marble to produce fully consistent and persistent 3D scenes, which could be valuable for on-the-fly game content or reliable robot training data.
Academia and industry alike are testing new ways to blend AI with 3D worlds. Stanford’s 3D Generalist uses a large language model to decide how to modify scenes by adding new objects, while Princeton researchers are exploring scene generation through code to empower LLMs to generate and interact with environments. Google DeepMind’s SIMA 2 demonstrates how AI agents can reason, learn, and collaborate within virtual 3D worlds, opening doors to new gameplay paradigms.
As 3D-capable AI becomes more central to development, Tencent could play a pivotal role among a growing cadre of Chinese AI firms eyeing this space. Beyond gaming, Tencent’s WeChat ecosystem—an everything-in-one platform in China—gives the company a unique distribution and integration advantage, with YuanBao serving as a built-in chatbot. The combination of gaming expertise and a broad ecosystem could provide Tencent with a distinctive edge in a 3D AI-driven future.
This overview comes from Will Knight’s AI Lab newsletter, offering a snapshot of where 3D AI in games stands today and where it might head next. For ongoing coverage of AI in the lab and industry, see Knight’s newsletters and related reporting.
