The repository provides official implementations of StyleKeeper and VSP, two approaches for controlling visual style in text-to-image diffusion models without requiring fine-tuning. Both methods operate during the denoising process by swapping key and value projections in self-attention layers to transfer style from reference images while preventing content leakage. Published at ICCV 2025 with VSP winning Best and Oral recognition at CVPRW AI4CC 2024.