This project is an experiment which explores the capabilities of machine learning for generating new patterns based on input data from 18th-century church fabric archives from Brno. The neural network uses these motifs to generate a wide array of possibilities aiming to produce the most reliable imitations of the original patterns by successive iterations. The created patterns are projected onto a screen as a sequence with a smooth transformation of motifs. They are also embroidered on canvas to symbolically link the physical and virtual reality. The preparation of the dataset is an authorial element, which has a major influence on the final form of the project. The work emphasises the significance of human decisions behind the seemingly automated process.