Tencent has launched HY-World 2.0, a revolutionary 3D world model that converts various inputs like text, images, and videos into editable, persistent 3D scenes. Unlike traditional models that generate pixel-based videos, HY-World 2.0 generates real 3D assets such as meshes and Gaussian splats, which can be directly imported into software like Blender and Unity. The innovative architecture includes a four-stage pipeline for panorama creation, camera trajectory planning, world expansion, and scene reconstruction, leading to more stable and controllable outputs. Additionally, the core technology, WorldMirror 2.0, has been open-sourced, allowing developers easy access to the reconstruction model and paving the way for further advancements in 3D world generation and interaction.

$TCEHY: TCEHY is the American Depositary Receipt ticker for Tencent Holdings on U.S. exchanges. It represents shares in the company behind AI innovations like HY-World 2.0.
Tencent: Tencent is a leading Chinese technology conglomerate focused on AI foundation models, gaming, cloud services, and social platforms. Its Hunyuan AI platform develops multimodal generative technologies including text-to-video and 3D content creation. In this news, Tencent released HY-World 2.0, a framework that advances 3D world generation and reconstruction from diverse inputs.
WorldNav: WorldNav is the spatial planning component in HY-World 2.0 that generates intelligent camera trajectories from panoramas, prioritizing meaningful paths with semantic understanding and collision avoidance. It ensures natural exploration coverage in generated worlds. Introduced in the HY-World 2.0 framework, its code release is upcoming.
HY-Pano 2.0: HY-Pano 2.0 is the panorama generation stage of HY-World 2.0, creating 360-degree views from text or single images using end-to-end implicit learning without camera metadata. It initializes expansive scene representations for downstream processing. As part of the HY-World 2.0 release, its model weights and code are slated for open-source soon.
HY-World 2.0: HY-World 2.0 is Tencent’s multimodal world model framework designed for generating navigable 3D worlds from text, images, or videos and reconstructing scenes from photos or footage. It produces editable 3D assets like meshes and Gaussian splats compatible with game engines such as Unity and Unreal. The release emphasizes a four-stage pipeline for superior 3D consistency and real-time interaction over traditional video models.3738
Tencent Hunyuan: Tencent Hunyuan is Tencent’s AI division developing foundational models for text, image, video, and 3D generation, including tools like HunyuanVideo and Hunyuan3D. It hosts repositories for advanced generative technologies on GitHub and Hugging Face. Tencent Hunyuan announced and partially open-sourced HY-World 2.0, positioning it as a state-of-the-art 3D world model.2728
WorldMirror 2.0: WorldMirror 2.0 is a unified feed-forward model in HY-World 2.0 that reconstructs 3D scenes from multi-view images or videos in one pass, outputting depth maps, normals, camera poses, point clouds, and Gaussian splats. It supports flexible resolutions and integrates geometric priors for high-fidelity digital twins. Tencent open-sourced its inference code and weights as part of the HY-World 2.0 launch.23
WorldStereo 2.0: WorldStereo 2.0 is the world expansion module of HY-World 2.0, synthesizing new scene views along planned trajectories with precise camera control and memory mechanisms for geometric consistency. It bridges initial panoramas to full 3D compositions. Featured in HY-World 2.0, it enhances novel view synthesis and is set for open-source.

`json
{
“3D Asset Output”: “Generates persistent, editable 3D assets such as meshes and Gaussian splats that are compatible with platforms like Blender, Unity, Unreal Engine, and Isaac Sim.”,
“Open-Source Release”: “WorldMirror 2.0 inference code and model weights are immediately available on GitHub and Hugging Face, with additional components of the HY-World 2.0 pipeline expected to be released in stages.”,
“Pipeline Innovation”: “Utilizes a multi-stage process comprising panorama creation, camera path planning, scene expansion, and 3D scene assembly to ensure stable and manageable 3D world creation.”
}
`