Tencent has significantly advanced the frontier of generative AI by releasing and open-sourcing HY-World 2.0. Moving beyond the generation of 2D images or short video clips, HY-World 2.0 is a multimodal artificial intelligence system capable of generating, reconstructing, and simulating fully navigable 3D environments from text, images, and video inputs.
This release marks a strategic shift for Tencent’s Hunyuan model lineup, firmly positioning it in the realm of production-ready 3D assets. The defining characteristic of HY-World 2.0 is its focus on developer usability. Unlike earlier experimental systems, this model generates editable 3D assets that support industry-standard formats, including meshes, 3D Gaussian splatting, and point clouds.
Crucially, these outputs are designed to be exported directly into major game engines like Unity and Unreal Engine, as well as other professional graphics pipelines.
A Four-Stage Generation Pipeline
HY-World 2.0 operates through a sophisticated four-stage pipeline designed to ensure spatial coherence and visual fidelity:
1.HY-Pano-2.0: This panorama model first generates a complete 360-degree scene representation from standard 2D images. It achieves this without requiring camera metadata, utilizing a blend of real and synthetic training data.
2.WorldNav: A trajectory planning module that determines how to move through the generated environment. It uses spatial reasoning and game-style pathfinding to create coherent exploration paths, actively avoiding errors like clipping through walls.
3.WorldStereo 2.0: This module expands the environment along the paths determined by WorldNav, maintaining visual detail and spatial consistency as new areas of the world are generated.
4.WorldMirror 2.0: The final stage integrates all outputs into a unified 3D environment, producing the exportable assets (meshes, Gaussian splats) ready for use in external software.
Targeting the Gaming Industry
Tencent’s emphasis on game development workflows is highly strategic. Through Tencent Games and its massive network of studios (including TiMi Studio Group and LightSpeed Studios), the company operates one of the largest game production ecosystems globally.
HY-World 2.0 is explicitly geared toward real-world workflows in this sector. Development teams can use the generated environments for rapid map building, level prototyping, and asset pre-production. By automating the creation of base environments, designers can generate early-stage layouts in minutes before manually refining them, drastically reducing the cost and time required to build complex 3D worlds.
The rapid iteration of the HY-World series, from version 1.0 in July 2025 to version 2.0 today, reflects a broader trend in generative AI: the move from generating media to generating structured, interactive environments that serve as building blocks for complex software systems. With HY-World 2.0, Tencent is positioning itself as a foundational provider of the tools that will build the next generation of digital spaces.