The paper introduces LaRE², a novel method for detecting diffusion-generated images. It addresses the privacy and security concerns arising from the improved image generation quality of diffusion models. LaRE² consists of two main components: Latent Reconstruction Error (LaRE) and Error-Guided Feature Refinement (EGRE). LaRE is a reconstruction error-based feature extracted in the latent space, which is more efficient and faster than existing methods. EGRE refines the image features guided by LaRE, enhancing their discriminative power. The method is evaluated on the GenImage benchmark, achieving significant improvements in accuracy and average precision compared to state-of-the-art methods. Experiments demonstrate that LaRE² is effective and generalizable, with a speed enhancement of 8 times in feature extraction.The paper introduces LaRE², a novel method for detecting diffusion-generated images. It addresses the privacy and security concerns arising from the improved image generation quality of diffusion models. LaRE² consists of two main components: Latent Reconstruction Error (LaRE) and Error-Guided Feature Refinement (EGRE). LaRE is a reconstruction error-based feature extracted in the latent space, which is more efficient and faster than existing methods. EGRE refines the image features guided by LaRE, enhancing their discriminative power. The method is evaluated on the GenImage benchmark, achieving significant improvements in accuracy and average precision compared to state-of-the-art methods. Experiments demonstrate that LaRE² is effective and generalizable, with a speed enhancement of 8 times in feature extraction.