DreamCraft3D++: Efficient Hierarchical 3D Generation with Multi-Plane Reconstruction Model

TL;DR

DreamCraft3D++ is introduced, an extension of DreamCraft3D that enables efficient high-quality generation of complex 3D assets and replaces the time-consuming geometry sculpting optimization with a feed-forward multi-plane based reconstruction model, speeding up the process by 1000x.

Abstract

We introduce DreamCraft3D++, an extension of DreamCraft3D that enables efficient high-quality generation of complex 3D assets. DreamCraft3D++ inherits the multi-stage generation process of DreamCraft3D, but replaces the time-consuming geometry sculpting optimization with a feed-forward multi-plane based reconstruction model, speeding up the process by 1000x. For texture refinement, we propose a training-free IP-Adapter module that is conditioned on the enhanced multi-view images to enhance texture and geometry consistency, providing a 4x faster alternative to DreamCraft3D's DreamBooth fine-tuning. Experiments on diverse datasets demonstrate DreamCraft3D++'s ability to generate creative 3D assets with intricate geometry and realistic 360° textures, outperforming state-of-the-art image-to-3D methods in quality and speed. The full implementation will be open-sourced to enable new possibilities in 3D content creation.

Figures (6)

• Figure 1: By lifting 2D images to 3D, DreamCraft3D++ achieves 3D generation with rich details and holistic 3D consistency. Please refer to the demo video for more results.
• Figure 2: DreamCraft3D++ pipeline. A single input image is processed by multi-view diffusion models to generate orthogonal, consistent views and normal maps. A feed-forward sparse-view 3D reconstruction model (Sec. \ref{['sec:lrm']}) infers textured meshes from the multi-view images using a convolutional U-Net to map input to non-orthogonal planes, decoded into Flexicubes. Finally, a training-free object-aware diffusion prior enhances high-frequency geometry and texture details via score distillation (Sec. \ref{['sec:refiner']}).
• Figure 3: Qualitative comparison with baselines on the GSO dataset.
• Figure 4: Qualitative comparison with baselines on the Internet images.
• Figure 5: Ablation study on the strategies of the refinement.