Chord: Chain of Rendering Decomposition for PBR Material Estimation from Generated Texture images

Ubisoft La Forge1, ETH Zürich2
SIGGRAPH Asia 2025

*Equal Contribution

PBR materials generated by our method.

Abstract

In this paper we propose a generate-and-estimate framework to produce high-quality SVBRDF channels for physically-based rendering of materials. In the generation stage, we leverage the diverse conditioning techniques of text-to-image models to synthesize tileable texture images with creative control. In the estimation stage, we introduce a chain-of-rendering-decomposition (Chord) scheme, which sequentially predicts SVBRDF channels by feeding previously extracted representations into a single-step, image-conditioned diffusion model. Our material estimation method demonstrates strong robustness on both generated textures and in-the-wild photographs. Moreover, we showcase the flexibility of our entire framework across diverse applications, including text-to-material, image-to-material, structure-guided generation, and material editing.

Method Overview

Stage 1: Tileable texture image ($ I_\text{RGB} $) generation using a fine-tuned diffusion model, controllable via user guidance (text prompts, reference images, or other control types).

Stage 2: Material estimation predicts SVBRDF channels sequentially:

  1. Basecolor $\hat{b}$.
  2. Normal $\hat{n}$ (with height $\hat{h}$ derived via normal integration).
  3. Roughness $\hat{r}$ and metalness $\hat{m}$.
Each step's input is computed from previous outputs. LEGO-conditioning provides modality-specific weights while maintaining shared backbone weights for alignment. Details of LEGO-conditioning are provided in Section 4.2.1.

Material Estimation Comparisons



$\dagger$: re-trained on our dataset, $\ast$: author-provided weights.

Applications



Oak Tree Texture

Text to Material

Image to Material

Structure-controlled Generation

Before
+Lava

Material Editing

Before
Edited

Adding 🍌

Before
+Banana

BibTeX

@misc{ying2025chord,
      title={Chord: Chain of Rendering Decomposition for PBR Material Estimation from Generated Texture Images}, 
      author={Zhi Ying and Boxiang Rong and Jingyu Wang and Maoyuan Xu},
      year={2025},
      eprint={2509.09952},
      archivePrefix={arXiv},
      primaryClass={cs.GR},
      url={https://arxiv.org/abs/2509.09952}, 
}