Textrix | Native 3D Texture Generation via Sparse Attribute Grids

TEXTRIX: Latent Attribute Grid for Native Texture Generation and Beyond

Yifei Zeng^1,2,* Yajie Bao^1,2,* Jiachen Qian^3,2 Shuang Wu^1,2 Youtian Lin¹ Hao Zhu¹ Buyu Li⁴ Feihu Zhang² Xun Cao¹ Yao Yao^1,†

¹Nanjing University ²DreamTech ³HKU ⁴OriginArk

^*Equal contribution ^†Corresponding author

Paper arXiv Code

TEXTRIX is a native 3D texturing framework that generates high-fidelity textures and precise 3D segmentations by leveraging a sparse attribute grid, fundamentally resolving inter-view inconsistencies and enabling seamless texture synthesis.

Traditional 3D texture generation methods suffer from inter-view inconsistencies and incomplete surface coverage. TEXTRIX introduces a native 3D attribute generation framework using a latent 3D attribute grid and sparse attention-based Diffusion Transformer, enabling direct volumetric coloring without multi-view fusion limitations.

Native 3D Texturing

Sparse attribute grid eliminates inter-view inconsistencies and seam artifacts, providing seamless texture generation across all viewpoints.

Spatial Aware Conditioning

Novel visual information projection ensures high-fidelity detail preservation and accurate texture-to-geometry alignment.

Unified Architecture

Single framework handles texture synthesis, 3D segmentation, and beyond with state-of-the-art performance across tasks.

Interactive Model Gallery

Click any model to explore in detail

Method Overview

TEXTRIX VAE Architecture: We introduce a native latent 3D attribute representation that contains properties such as color, semantic labels, and PBR materials within each sparse voxel. An end-to-end attribute VAE encodes this representation into a continuous and compact latent space.

TEXTRIX Diffusion Transformer (DiT): This image-conditioned model operates on sparse latents to perform unified generation (texturing) and perception (segmentation). We introduce a novel sparse latent conditioning strategy that ensures high-fidelity alignment with the input image.

Comparison & Results

Single-view Comparison

TexGen

Paint3D

TRELLIS

Ours

Demonstration of TEXTRIX generating texture from a single input view.

Multi-view Comparison

Comparison across multiple views demonstrating seam-free texture generation.

3D Segmentation Comparison

SAMesh

SAMPart3D

PartField

Ours

3D segmentation comparison showing TEXTRIX's ability to generate accurate segmentations on complex geometry.

PBR Results

Barrel

Chair

Gun

Gun 2

Head

Launcher

Pot

Robot

PBR material results showing TEXTRIX's ability to generate physically-based rendering materials with realistic lighting and surface properties.

BibTeX

@article{textrix2025,
  title     = {TEXTRIX: Latent Attribute Grid for Native Texture Generation and Beyond},
  author    = {Yifei Zeng and Yajie Bao and Jiachen Qian and Shuang Wu and Youtian Lin and Hao Zhu and Buyu Li and Feihu Zhang and Xun Cao and Yao Yao},
  journal   = {arXiv preprint arXiv:2512.02993},
  year      = {2025},
}