This repository implements the Plug-and-Play Diffusion Features method from CVPR 2023, enabling text-driven image-to-image translation by extracting and re-injecting intermediate features from a pretrained diffusion model. The approach allows users to control image synthesis by manipulating spatial features while preserving the generative model’s overall structure. It builds on Stable Diffusion v1.4 and includes feature extraction pipelines and inference code for image translation tasks.