The Bitmap Font Generator is a deliberately reduced Variational Autoencoder developed to investigate generative type design, inspired by early digital tools such as the Font Editor for bitmap fonts. The glyphs are constructed on a strictly grid-based system and encoded in binary, making their structure and data representation directly comprehensible. The model was trained on a self-created, standardized dataset of 312 variants of the 26 uppercase letters. The VAE produces mostly legible glyphs while preserving fundamental formal principles, though occasional glitches reveal both the system’s limitations and its creative potential. Overall, the project demonstrates how technical constraints can make generative processes controllable and open up new perspectives on nonlinear, latent-space- based type variation.