A single simple patch can effectively detect AI-generated images. The paper proposes a method called SSP (Single Simple Patch) network, which uses the noise pattern of a single simple patch to distinguish between real and fake images. The method first extracts a simple patch from the image, then applies SRM filters to extract its noise pattern, which is used by a binary classifier to detect fake images. The SSP network is designed to generalize well across different generators. However, low-quality generated images can affect detection performance, so an enhancement module and a perception module are introduced to improve the quality of the patch and reduce the impact of artifacts like blur and compression. The perception module evaluates the probability of the input patch being blurry or compressed, and the enhancement module uses this information to improve the patch quality. The SSP network is tested on two benchmark datasets, GenImage and ForenSynths, and achieves state-of-the-art performance. The method is effective in detecting AI-generated images, even when the quality of the generated images is low. The results show that the SSP network outperforms existing methods in terms of accuracy and mAP. The method is robust to blur and compression, and the enhancement and perception modules significantly improve detection performance on low-quality images. The paper also conducts ablation studies to validate the effectiveness of individual components of the method. The results show that the SSP network is effective in detecting AI-generated images, and the enhancement and perception modules are crucial for improving detection performance on low-quality images. The method is simple, effective, and can be applied to a wide range of generators.A single simple patch can effectively detect AI-generated images. The paper proposes a method called SSP (Single Simple Patch) network, which uses the noise pattern of a single simple patch to distinguish between real and fake images. The method first extracts a simple patch from the image, then applies SRM filters to extract its noise pattern, which is used by a binary classifier to detect fake images. The SSP network is designed to generalize well across different generators. However, low-quality generated images can affect detection performance, so an enhancement module and a perception module are introduced to improve the quality of the patch and reduce the impact of artifacts like blur and compression. The perception module evaluates the probability of the input patch being blurry or compressed, and the enhancement module uses this information to improve the patch quality. The SSP network is tested on two benchmark datasets, GenImage and ForenSynths, and achieves state-of-the-art performance. The method is effective in detecting AI-generated images, even when the quality of the generated images is low. The results show that the SSP network outperforms existing methods in terms of accuracy and mAP. The method is robust to blur and compression, and the enhancement and perception modules significantly improve detection performance on low-quality images. The paper also conducts ablation studies to validate the effectiveness of individual components of the method. The results show that the SSP network is effective in detecting AI-generated images, and the enhancement and perception modules are crucial for improving detection performance on low-quality images. The method is simple, effective, and can be applied to a wide range of generators.