Hunyuan3D-Buffalo 1.0: A Unified Multimodal Model for Scalable 3D Generation, Understanding, and Editing

2026-08-03

AI Summary

Existing unified 3D multimodal models are hindered by the scarcity and geometric inconsistency of available multimodal data. This work proposes the first unified framework that jointly addresses 3D understanding, text-to-3D generation, instruction-guided editing, and part-level generation, enabled by scalable joint training on a large-scale, geometrically consistent 3D multimodal corpus comprising 87 million samples. The approach integrates Hunyuan3D-VLM for semantic and spatial reasoning with Hunyuan3D DiT for high-fidelity synthesis, and introduces source object representations into the diffusion process to preserve structural consistency. The resulting model achieves state-of-the-art performance on text-to-3D generation and 3D editing benchmarks while demonstrating strong 3D comprehension and fine-grained part generation capabilities.

Abstract

Recent advances in image generation have demonstrated the potential of unified multimodal models that integrate understanding, generation, and editing. However, unified 3D modeling remains constrained by scarce multimodal data, particularly the lack of large-scale and geometrically consistent editing data. To address this limitation, we propose Hunyuan3D-Buffalo 1.0, a unified framework supporting 3D understanding, text-to-3D generation, instruction-guided 3D editing, and text-grounded part generation within a single architecture. To enable scalable training, we construct an 87M-scale 3D multimodal corpus, comprising 25M understanding samples, 50M text-to-3D pairs, and 12M editing pairs generated using Nano3D-v2. Architecturally, the framework 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. Our analysis further shows that both generation and understanding improve editing, demonstrating the effectiveness of unified 3D multimodal training. Project Page: https://tencent-hunyuan.github.io/Hunyuan3D-Buffalo1.0/

Problem

• 3D generation
• 3D editing
• multimodal learning
• unified model
• 3D understanding

Innovation

• unified multimodal model
• scalable 3D generation
• instruction-guided 3D editing
• 3D part generation
• multimodal corpus