VideoNeuMat: Neural Material Extraction from Generative Video Models

SIGGRAPH 2026

1University of Manchester, 2NVIDIA, 3University of California Santa Barbara
*Equal contribution
VideoNeuMat representative neural material results.

VideoNeuMat generates realistic and diverse neural materials from text or image prompts.

Abstract

Creating photorealistic materials for 3D rendering requires exceptional artistic skill. Generative models for materials could help, but are currently limited by the lack of high-quality training data. While recent video generative models effortlessly produce realistic material appearances, this knowledge remains entangled with geometry and lighting.

We present VideoNeuMat, a two-stage pipeline that extracts reusable neural material assets from video diffusion models. First, we finetune a large video model to generate material sample videos under controlled camera and lighting trajectories, effectively creating a virtual gonioreflectometer that preserves the model's material realism while learning a structured measurement pattern.

Second, we reconstruct compact neural materials from these videos through a Large Reconstruction Model. From generated video frames, our model predicts neural material parameters that generalize to novel viewing and lighting conditions.

Pipeline

Our method has two stages. First, we finetune a video diffusion model into a virtual gonioreflectometer that generates structured material videos from text or image prompts. Second, a feed-forward LRM infers a NeuMIP-style material from 17 frames using a rendering loss under novel views and lights. The resulting material supports relighting and novel shapes.

The two-stage VideoNeuMat pipeline from the paper.

Results

Text to Material

Basketweave reconstruction. Basketweave on curved geometry.
Basketweave
Black cow fur reconstruction. Black cow fur on curved geometry.
Black cow fur
Brown woven leather reconstruction. Brown woven leather on curved geometry.
Brown woven leather
Crimson velvet reconstruction. Crimson velvet on curved geometry.
Crimson velvet
Dragon carved wood reconstruction. Dragon carved wood on curved geometry.
Dragon carved wood
Frost ivy reconstruction. Frost ivy on curved geometry.
Frost ivy
Knurled metal reconstruction. Knurled metal on curved geometry.
Knurled metal
Pizza crust reconstruction. Pizza crust on curved geometry.
Pizza crust
Stone wall reconstruction. Stone wall on curved geometry.
Stone wall
White faux fur reconstruction. White faux fur on curved geometry.
White faux fur
Hay reconstruction. Hay on curved geometry.
Hay
Honeycomb hives reconstruction. Honeycomb hives on curved geometry.
Honeycomb hives

Single Image to Material

Input Ours Full GT
Sci-fi panel input. VideoNeuMat full result for sci-fi panel. Ground truth sci-fi panel.
Input Ours Full GT
Rubble wall input. VideoNeuMat full result for rubble wall. Ground truth rubble wall.

BibTeX

@inproceedings{xue2026videoneumat,
  author  = {Xue, Bowen and Hadadan, Saeed and Zeng, Zheng and Rousselle, Fabrice and Montazeri, Zahra and Hasan, Milos},
  title   = {VideoNeuMat: Neural Material Extraction from Generative Video Models},
  booktitle = {ACM SIGGRAPH 2026 Conference Papers},
  year    = {2026},
}