Tencent Open-Sources HY-World 2.0 for 3D World Generation and Reconstruction

Tencent has released HY-World 2.0, a multimodal world model that can generate, reconstruct, and simulate interactive 3D environments from text, images, and videos, according to details on Hugging Face. The company has open-sourced the model’s technical report, partial code, and model weights, with additional components such as full inference code and supporting modules to be released soon.

HY-World 2.0 can create persistent, editable 3D assets such as meshes, Gaussian splattings, and point clouds, which can be imported directly into engines like Unity, Unreal Engine, and Isaac Sim. The framework includes tools for both world generation—from text or single images—and reconstruction from multi-view images or videos.

The system integrates several components: HY-Pano 2.0 for panorama generation, WorldNav for trajectory planning, WorldStereo 2.0 for scene expansion, and WorldMirror 2.0 for world composition and reconstruction. The open-source release includes the WorldMirror 2.0 model and code, with a Gradio demo for real-time visualization of 3D outputs.

HY-World 2.0 builds on previous Hunyuan world modeling research and aims to provide an open, reproducible platform for the creation of interactive 3D worlds across gaming, simulation, and digital content applications.