Texture Matching Lite
Pick a folder of reference textures, then pick an image to match. The model runs entirely in your browser, so nothing is uploaded anywhere.
This is the lightweight counterpart to the Texturescope Toolbox. It uses the Asilomar triplet-loss model, trained on 64×64 texture tiles and converted to TensorFlow.js. Unlike the Toolbox, it needs no server, so it still works. The original training notebook is on Google Colab.
1. Reference folder
Every image in the folder becomes part of the reference set. Tiles exported from the Image Processing page work well here.