Controllable Stable Diffusion Using ControlNet for Stylized Javanese Script Typography Synthesis

Abstract

Preserving traditional indigenous scripts in digital environments requires robust computational frameworks that maintain strict orthographic fidelity alongside flexible artistic expression. Indigenous writing systems possess intricate non-linear structures, vertical character stacking, and delicate diacritics that frequently trigger severe structural distortion and character hallucinations in standard text-to-image generative architectures. This research aims to develop an automated hybrid pipeline that unites rule-based natural language processing with probabilistic generative artificial intelligence to synthesize high-fidelity stylized Javanese script calligraphy. The proposed method integrates linguistic normalization and deterministic transliteration to produce standardized visual guidance masks, followed by dual spatial conditioning that executes edge boundary extraction and depth estimation in parallel. Furthermore, low-rank adaptation modules are embedded into the latent diffusion process to extract complex visual material textures from curated paired synthetic datasets. Experimental evaluations demonstrate that the optimal parameter configuration successfully synthesizes five distinct aesthetic material styles with zero typographical errors and zero percent physical glyph deformation. The dual spatial constraints effectively lock vertical stacking components and microscopic diacritics while rendering realistic three-dimensional relief depth and specular surface lighting. In conclusion, the proposed hybrid architecture resolves structural disorientations in baseline diffusion models, providing a reliable technological infrastructure for digital cultural preservation, automated graphic typography production, and creative ethnic heritage dissemination.