Diffusion-Driven Image Generation with Structural Precision and Style Fidelity
Published in Manuscript in Preparation, 2025
A one-shot denoising diffusion framework for high-fidelity multilingual font generation. The model fuses multi-scale style vectors at multiple U-Net stages, a Sobel-based structural consistency loss, and CLIP-guided style alignment to produce glyphs with stroke-level precision across Korean, Chinese, and Latin scripts.
Recommended citation: Abdul Sami, Jaeyong Choi. (2025). "Diffusion-Driven Image Generation with Structural Precision and Style Fidelity." Manuscript in Preparation.