Abstract
This study mainly introduces a method combining the Stable Diffusion Model(SDM) and Parameter-Efficient Fine-Tuning method for generating ChineseLandscape Paintings. This training process is accelerated by combining LoRAwith pre-trained SDM and DreamBooth with pre-trained SDM, respectively. On theChinese Landscape Paintings Internet dataset used in this paper, this studyfinds that SDM combined with DreamBooth exhibits superior performance,outperforming other models, including the generic pre-trained SDM andLoRA-based fine-tuning SDM. The SDM combined with DreamBooth achieves a FID of12.75 on the dataset and outperforms all other models in terms of expertevaluation, highlighting the model's versatility in the field of ChineseLandscape Paintings given the unique identifier, high fidelity and highquality. This study illustrates the potential of specialised fine-tuning methodto improve the performance of SDM on domain-specific tasks, particularly in thedomain of Landscape Paintings.