关键词: 扩散模型, 图像编辑, 低秩适用, 模型微调, U-net

Abstract: Addressing the issues of task singularity, user-unfriendliness, and low fidelity in current mainstream image editing
methods, a diffusion model-based method for high-fidelity image editing was proposed. The method utilized the prevalent stable
diffusion model as the backbone network, initially fine-tuned using the Low Rank Adaptation (LoRA) method to better reconstruct the
original image. Subsequently, the refined model was employed to infer images with simple prompts through a designed framework,
ultimately generating edited images. Furthermore, a dual-layer U-net structure was extended from the aforementioned method for
specific image editing tasks and exploration as video synthesis. Comparative experiments with leading methods Imagic, DiffEdit, and
InstructPix2Pix on the Tedbench dataset demonstrated the method's capability to perform various editing tasks, including non-rigid
editing, with strong editability. The method exhibited a 30.27% decrease in Learned Perceptual Image Patch Similarity (LPIPS) index
compared to Imagic, indicating higher fidelity.

Key words: Diffusion model, Image editing, Low-Rank Adaptation (LoRA), Model fine-tuning, U-net