Hunyuan3D-Omni

Why we included this project

Teams building 3D content pipelines will find this a practical reference for controllable asset generation. Built on the Hunyuan3D 2.1 architecture, it adds a unified control encoder that lets you steer generation from several input signals at once, including point clouds, voxel grids, skeletal poses, and 3D bounding boxes, rather than relying on a single conditioning type. That flexibility matters when you need to fit generated geometry into an existing scene, match a specific human pose, or respect spatial constraints from a layout. The project ships pretrained models and runs in about 10 GB of VRAM, which makes it approachable for a single workstation or a small GPU box. It is a research-oriented framework, so expect to work with the code and model weights rather than a turnkey application, but the multi-modal conditioning approach is a useful template if you are exploring controllable 3D generation.