- Abstract: Publicly available labelled handwriting data is fairly limited in its representation of styles as well as in the volume of examples for many topics. We find that trying to use these publically available datasets as training data on unrelated unlabelled handwriting datasets produces unsatisfactory results and would not produce a trained system capable of performing adequately in real world tasks. We propose a method of character and word generation using fonts as templates, as large numbers of handwriting fonts are available for personal use online. Our technique, based on modifying previous work in mechanical handwriting modeling and template based generation and extending that to arbitrary images of letter and words with an automatic method of generating templates through splinification. We find that we get reasonable results on MNIST, EMNIST, IAM, and a proprietary dataset created from Boeing aircraft maintenance forms when no training data is available. This method requires minimal training and generates in a fast, easily parallelizable fashion.