Research on Generating Cultural Relic Images Based on a Low-Rank Adaptive Diffusion Model
Juntao Deng, Xu Cao, Bingqi Cheng · 2024
In the field of artificial intelligence, large generative models like Stable Diffusion have made strides in image generation. However, they struggle to accurately generate images of cultural relics with specific historical features. This study uses Low-Rank Adaptive (LoRA) fine-tuning to optimize the Stable Diffusion model for this purpose. We collected and organized cultural relic images and descriptions to customize the model, allowing effective fine-tuning while maintaining its stability. Experimental results show that the fine-tuned model accurately generates images with historical characteristics, aligning with historical data and expert evaluations. We also explore potential applications in cultural creation and artifact restoration, aiming to inspire interdisciplinary innovations and collaborations in AI applications for the cultural sector.