Advancements in Text-to-3D Generation

The field of text-to-3D generation is moving towards more sophisticated and realistic outputs, with a focus on incorporating 3D priors and structure-aware modeling. Recent developments have enabled the generation of complex structures, such as bonsai, and have improved the consistency of multi-view rendering. Additionally, there is a growing emphasis on texture generation and editing capabilities, allowing for more detailed and realistic outputs. Notable papers in this area include:

  • ORIGEN, which introduces a zero-shot method for 3D orientation grounding in text-to-image generation, and
  • IntrinsiX, which generates high-quality physically-based rendering maps from text descriptions,
  • 3DBonsai, which proposes a novel text-to-3D framework for generating 3D bonsai with complex structures,
  • ConsDreamer, which mitigates view bias in zero-shot text-to-3D generation, and
  • MD-ProjTex, which achieves fast and consistent text-guided texture generation for 3D shapes.

Sources

ORIGEN: Zero-Shot 3D Orientation Grounding in Text-to-Image Generation

IntrinsiX: High-Quality PBR Generation using Image Priors

3DBonsai: Structure-Aware Bonsai Modeling Using Conditioned 3D Gaussian Splatting

ConsDreamer: Advancing Multi-View Consistency for Zero-Shot Text-to-3D Generation

MD-ProjTex: Texturing 3D Shapes with Multi-Diffusion Projection

Built with on top of