AEROBLADE is a training-free method for detecting images generated by latent diffusion models (LDMs) using autoencoder (AE) reconstruction error. The method leverages the inherent property of LDMs that generated images can be more accurately reconstructed by the AE than real images. By computing the reconstruction error between an image and its reconstruction through the AE's encoder and decoder, AEROBLADE can effectively distinguish real images from generated ones. This approach is simple to implement and does not require any training, yet it achieves performance comparable to trained detectors. AEROBLADE is effective against state-of-the-art LDMs such as Stable Diffusion and Midjourney. Beyond detection, the method allows for qualitative analysis of images, which can be used to identify inpainted regions. The method is evaluated on various datasets and shows high detection accuracy, with a mean average precision (AP) of 0.992 on several models. AEROBLADE also provides rich qualitative information, giving insights into how well certain regions can be reconstructed. The method is implemented and made available at https://github.com/jonasricker/aeroblade.AEROBLADE is a training-free method for detecting images generated by latent diffusion models (LDMs) using autoencoder (AE) reconstruction error. The method leverages the inherent property of LDMs that generated images can be more accurately reconstructed by the AE than real images. By computing the reconstruction error between an image and its reconstruction through the AE's encoder and decoder, AEROBLADE can effectively distinguish real images from generated ones. This approach is simple to implement and does not require any training, yet it achieves performance comparable to trained detectors. AEROBLADE is effective against state-of-the-art LDMs such as Stable Diffusion and Midjourney. Beyond detection, the method allows for qualitative analysis of images, which can be used to identify inpainted regions. The method is evaluated on various datasets and shows high detection accuracy, with a mean average precision (AP) of 0.992 on several models. AEROBLADE also provides rich qualitative information, giving insights into how well certain regions can be reconstructed. The method is implemented and made available at https://github.com/jonasricker/aeroblade.