Hunyuan World 1.1 (WorldMirror) training code goes public

Tencent's WorldMirror training-code release gives researchers a way to adapt its image-and-video reconstruction model. WorldMirror estimates 3D scene information, including depth, camera properties and points in space. The training documentation provides dataset preparation, training configurations and checkpoint evaluation.

Researchers can choose which kinds of output the model learns to predict, such as depth or camera information, by enabling or disabling the corresponding prediction components. That makes the release useful for experiments requiring a particular reconstruction task, rather than only running the supplied demonstration.