Synthetic Image Generation with Stable Diffusion and Trained LoRA Model for Industrial Areas

The rapid advancement of technology has significantly influenced synthetic image generation methodologies, widely applied in fields such as healthcare, manufacturing, and automative. This study explores the integration of Stable Diffusion models and Low-Rank Adaptation (LoRA) techniques to generate high-quality synthetic images, aiming to overcome the limitations of high costs and the need for extensive annotated datasets in industrial applications. We employed the Stable Diffusion model for its precision and customizability, enhanced by a LoRA model trained on Otokar quality inspection datasets. This combination facilitates faster, high-quality image generation with minimal data. The quality of the generated image was evaluated using various metrics such as SSIM, PSNR and MSE. The results show that the synthetically generated images are of sufficient quality and will overcome the limitations of traditional data collection methods, especially those involved in inspection systems. In the future, we will extend the application to other areas, incorporating advanced generative models, and developing user-friendly platform for seamless industrial integration.