Hunyuan3D 2.0 Mini Turbo is a high-speed generative model developed by Tencent designed for the rapid synthesis of 3D assets from single images or text prompts. As a distilled, lightweight variant of the Hunyuan3D 2.0 family, it is optimized for consumer-grade hardware and interactive applications. The model achieves a significant performance leap over previous iterations, reportedly providing over a 30x speedup in shape generation, enabling the creation of basic 3D geometry in approximately one second.
The model's architecture utilizes a two-stage generation pipeline that decouples shape and texture synthesis. The first stage employs Hunyuan3D-DiT, a flow-based Diffusion Transformer (DiT) that reconstructs untextured 3D meshes. The second stage uses Hunyuan3D-Paint to generate high-resolution, vivid texture maps. This approach allows the model to produce detailed geometric structures with production-ready textures while maintaining alignment with the input conditions.
Technical specifications for the Mini Turbo version include a parameter count of approximately 0.6 billion. It incorporates advanced optimization techniques such as Step Distillation (or Guidance Distillation), which reduces the number of inference steps required to produce high-fidelity results. This efficiency allows the model to operate within a low memory footprint, requiring as little as 5GB of VRAM for shape generation and approximately 12GB for the full shape-and-texture workflow.
Output from the model typically includes industry-standard formats like .obj and .glb, making it compatible with major 3D software and game engines. For best results, users are encouraged to use clear, centered images with neutral backgrounds, though the model's multi-view diffusion priors assist in inferring complex occluded geometries and maintaining consistent textures across different viewing angles.