Generation of Batik Cual Bangka Belitung Motif Variations Using Stable Diffusion Models

Abstract

Batik Cual Bangka Belitung is a visual cultural heritage characterized by repetitive motifs, fine line details, and high ornament complexity, presenting significant challenges for digital exploration without compromising its visual identity. This study aims to generate variations of Batik Cual motifs using a prompt-guided image-to-image (img2img) approach based on the Stable Diffusion model with a single-reference image. Furthermore, the research analyzes the influence of the strength parameter on the delicate balance between structural similarity and generative visual variation. The dataset consists of 13 independently collected Batik Cual motifs, pre-processed through RGB conversion and standardized to a resolution of 512×512 pixels. Controlled experiments were conducted using varied strength values of 0.4, 0.6, and 0.8, while maintaining other parameters constant. Quantitative evaluation utilized the Structural Similarity Index Measure (SSIM) to assess structural integrity and CLIP similarity to measure semantic alignment between the prompt and output image. The results indicate that increasing strength consistently decreases SSIM values, signifying greater structural deviation from the reference image, whereas CLIP similarity remains relatively stable across configurations. Quantitatively, a strength of 0.4 offers the optimal combination of structural similarity and semantic suitability. However, qualitative assessments reveal that for certain motifs, a strength of 0.6 produces a more balanced variation between pattern innovation and motif character preservation. These findings confirm a measurable trade-off between identity preservation and generative exploration, demonstrating the potential of Stable Diffusion as a controlled method for developing Batik Cual digital assets.