SUPRA-TEX: 3D Generation with Super-Aligned Texture
What it is
SUPRA-TEX is a method for single-image 3D generation that decouples visible-material estimation, surface alignment, and 3D completion. It uses an MLLM to predict PBR maps, projects them into latent space as fixed anchors, and employs an interleaved flow transformer to complete unobserved surfaces while preserving reference fidelity.
The problem it solves
Existing single-image 3D generation methods suffer from artifacts when fusing inconsistent multi-view predictions or fail to preserve fine visual details due to implicit conditioning.
What you could use it for
- Generating relightable 3D assets from a single reference image
- Texturing 3D meshes with high-fidelity materials that match visual details
- Creating coherent 3D models where visible regions match the input and hidden regions are plausibly completed
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.