PBR Texture Generator
Select input photos to be converted to textures. The data is processed locally and within seconds.
Please select input photos first..
About the pix2pbr Project
This demo is based on Thorben's master thesis: Synthesis of Textures for Automotive Simulations Using Deep Learning [Master Thesis]. Our neural networks were trained in PyTorch and exported to TensorFlow.js using onnx. They synthesize the most common PBR textures from the metallic roughness workflow based on a single RGB input image. Inputs for which the z-height correlates positively with the brightness are particularly suitable. If there are several input images, the textures are created for each image individually.