From Words to Wood: Text‐to‐Procedurally Generated Wood Materials

The Project In A Nutshell

This is the latest and most complete procedurally generated wood model worldwide as in 2024–2025.

• A procedurally generated wood model including growth rings, vessels/pores, knots, figure, and much more.
• An NLP based interface to control the procedural model using plain English text.

Keywords: Procedural generation of materials, NLP, Wood materials, Game-ready assets generation.

Acquired or Enhanced Skills

Project Problems and Solutions

This project began with a simple question: why isn’t there a wood model that brings all wood features together in a single system? Our initial answer was to build a highly complex material model that incorporated all required features for realistic wood. However, the final model had over 90 parameters, forcing new users to spend weeks just to learn which parameter controls which feature. To solve this, we used the most used communication tool in history, and that is English and built a natural-language interface that sits between the user and the procedural wood model.

This figure shows the pipeline of our system

Project Developement

Our procedurally generated wood model was build across different platforms, one of them is through the node graph system of the Blender software. Blender’s support for custom OSL nodes was especially valuable, this was helpfull in reduicing the amount of nodes needed to build the material generation system.

This figure shows the node graph for the procedurally generated wood model, note that it has been mostly build with custom nodes which are created using open shading language(OSL)

Results

Pine wood material with natural grain patterns

Detailed wood knot formation

Complex knot structure simulation

Ash wood with distinctive grain

Tree trunk cross-section with growth rings

Project teaser showcasing capabilities

Animation Demo

High-resolution animation demonstrating the material system in action

Acknowledgement

We express our sincere appreciation to Martin Mirbauer for his extensive support throughout the project and to Ali Ahmed Maher Dali for his expert assistance in data processing. This work received funding from the European Union’s Horizon 2020 research and innovation program under the Marie Sklodowska-Curie grant agreement No 956585, and from the Czech Science Foundation (GAČR) grant No. 22-22875S, which we gratefully acknowledge. The work was supported by the grant SVV-2025-260822. Open access publishing facilitated by Univerzita Karlova, as part of the Wiley – CzechELib agreement.