Research and Comparison of Ceramic Image Generation Through Fine-Tuning Stable Diffusion

This research explores the effectiveness of fine-tuning the Stable Diffusion model to generate ceramic images. We constructed a ceramic image-text pair dataset to fine-tune the model, to assess the effectiveness of the fine-tuning, We compared the generation results of the fine-tuned model with those of current mainstream generation models(such as DALL-E, Midjourney, and ERNIE-Bot), through both quantitative and qualitative evaluations. The quantitative assessment utilized CLIP Score and FID metrics, while the qualitative evaluation involved scoring and reviews from scholars in the field of ceramics. The experimental results showed that the fine-tuned model demonstrated significant improvements across all metrics, whereas other models exhibited their own advantages in different aspects. Our research provides new insights for applying image generation models in specific fields and offers valuable references for further optimizing model fine-tuning methods.