HY-World-2.0

Why we included this project

Tencent's Hunyuan lab releases HY-World 2.0 as a set of connected models rather than one tool. Together they take a real-world scene, rebuild it as an editable 3D representation, and then generate or simulate new worlds from it. Separate components handle scene reconstruction, panorama generation, and stereo imagery, so you are not locked into a single narrow task. Game designers, film previsualization artists, and embodied AI researchers who need consistent virtual environments from images will get the most immediate use out of it, since the repo ships inference code and model weights that run locally. The technical report also explains how these tasks share an architecture, which makes it a handy reference if you study world models. One caveat: the open-source plan is partial, so check which components are actually released before you build a production workflow around it.