Abstract

Recent advances in 3D-native generative models have accelerated assetcreation for games, film, and design. However, most methods still relyprimarily on image or text conditioning and lack fine-grained, cross-modalcontrols, which limits controllability and practical adoption. To address thisgap, we present Hunyuan3D-Omni, a unified framework for fine-grained,controllable 3D asset generation built on Hunyuan3D 2.1. In addition to images,Hunyuan3D-Omni accepts point clouds, voxels, bounding boxes, and skeletal posepriors as conditioning signals, enabling precise control over geometry,topology, and pose. Instead of separate heads for each modality, our modelunifies all signals in a single cross-modal architecture. We train with aprogressive, difficulty-aware sampling strategy that selects one controlmodality per example and biases sampling toward harder signals (e.g., skeletalpose) while downweighting easier ones (e.g., point clouds), encouraging robustmulti-modal fusion and graceful handling of missing inputs. Experiments showthat these additional controls improve generation accuracy, enablegeometry-aware transformations, and increase robustness for productionworkflows.