This study proposes a method for generating traditional Japanese architectural facades that preserve regional authenticity by incorporating chronological indicators—such as roofs, openings, and exterior wall types—and frontage width based on the traditional shakkanhō system into Pix2Pix training. Focusing on historic districts in northern Kyushu, annotated images were created to encode architectural elements and proportional rules, enabling the model to learn structural logic rather than relying solely on stylistic appearance. Generation experiments were conducted with and without these conditions, and outcomes were evaluated using SSIM and a Likert-based subjective assessment by participants familiar with the region’s architecture. Results demonstrate that embedding architectural knowledge significantly improves structural coherence, proportional accuracy, and cultural plausibility compared to conventional style-transfer-based generation. Even with limited training data, the proposed approach achieved regionally consistent and realistic facade outputs. This framework contributes to heritage preservation by supporting townscape analysis and the reconstruction of traditional building elevations in early design stages.