DragonDiffusion is a fine-grained image editing solution based on diffusion models, supporting various editing modes such as object movement, resizing, appearance replacement, and content dragging. It converts editing signals into gradients through feature correspondence loss and modifies the intermediate representation of the diffusion model. Feature correspondence loss considers semantic and geometric alignment at multiple scales and adds cross-branch self-attention to maintain consistency between the original image and the edited result.