SUPRA-TEX: 3D Generation with Super-Aligned Texture
SUPRA-Tex textures 3D meshes from one image using aligned PBR latents and reports commercial-level fidelity.
SUPRA-Tex generates relightable materials for a 3D mesh from one reference image by separating visible-material estimation, surface alignment, and observation-guided completion. An MLLM image editor predicts lighting-separated PBR maps, which are projected onto visible surfaces and encoded as fixed latents. An interleaved flow transformer completes unobserved regions while preserving those anchors. Trained only on public datasets, it reports state-of-the-art geometry-conditioned texturing and end-to-end asset generation, matching or exceeding leading commercial systems on in-the-wild images.
- SUPRA-Tex decouples PBR estimation, surface alignment, and 3D completion.
- An MLLM editor predicts lighting-separated PBR maps in the reference view.
- A flow transformer keeps observation latents fixed while completing surfaces.
- Public-data training matched or beat leading commercial texturing systems.
From the original article
1 min read at huggingface.co
Single image 3D generation aims to dress a given 3D mesh with relightable materials while faithfully preserving the visual details of a reference image. Existing methods either introduce artifacts when fusing inconsistent multi-view predictions or struggle to preserve fine details through implicit image-feature conditioning.
By Chen Yang, Jiemin Fang, Fanpeng Meng · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.