Researchers from Tencent have unveiled Hunyuan3D-Buffalo 1.0, a unified multimodal model that tackles multiple 3D tasks within a single architecture. The framework supports 3D understanding, text-to-3D generation, instruction-guided 3D editing, and text-grounded part generation, aiming to overcome the scarcity of large-scale 3D multimodal data.
To enable scalable training, the team constructed an 87M-scale 3D multimodal corpus, comprising 25 million understanding samples, 50 million text-to-3D pairs, and 12 million editing pairs generated using Nano3D-v2. This dataset addresses the lack of geometrically consistent editing data, a key bottleneck in unified 3D modeling.
Architecturally, the model combines Hunyuan3D-VLM for semantic, structural, and spatial understanding with Hunyuan3D DiT for high-fidelity 3D synthesis. The VLM provides multimodal semantic conditions for generation, while editing and part generation additionally condition the diffusion process on the source object representation to preserve its overall structure and unedited regions.
Extensive experiments show that Hunyuan3D-Buffalo 1.0 achieves state-of-the-art or leading performance on text-to-3D generation and 3D editing benchmarks, while exhibiting strong understanding and part-generation capabilities. The analysis further demonstrates that both generation and understanding improve editing, validating the effectiveness of unified 3D multimodal training.