This paper revisits Tree-Ring Watermarking, a recent diffusion model watermarking method known for its robustness against attacks. The authors investigate the method and find that its robustness is not only due to the carefully designed tree-ring pattern but also due to an unintentional distribution shift introduced during watermarking. This shift significantly contributes to the method's robustness against certain transformations, such as rotation and cropping/scale. However, the distribution shift does not help in multi-key identification, where the method struggles to distinguish between different keys, showing poor accuracy under various attacks. To address these issues, the authors propose RingID, a novel multi-channel heterogeneous watermarking approach that combines the advantages of different watermark types. RingID introduces several enhancements, including discretization, lossless imprinting, and spatial shifting, which significantly improve the method's ability to identify multiple keys. RingID achieves substantial improvements in both verification and identification tasks, with verification AUC improving from 0.975 to 0.995 and identification accuracy rising from 0.07 to 0.82. The method is robust to various attacks, including rotation and cropping/scale, and maintains high generation quality. The paper also discusses the design of RingID, including its ability to handle rotation, improve distinguishability through discretization, and increase capacity by using multiple channels. The results show that RingID outperforms Tree-Ring in multi-key identification, demonstrating its effectiveness in enhancing watermark identification capabilities.This paper revisits Tree-Ring Watermarking, a recent diffusion model watermarking method known for its robustness against attacks. The authors investigate the method and find that its robustness is not only due to the carefully designed tree-ring pattern but also due to an unintentional distribution shift introduced during watermarking. This shift significantly contributes to the method's robustness against certain transformations, such as rotation and cropping/scale. However, the distribution shift does not help in multi-key identification, where the method struggles to distinguish between different keys, showing poor accuracy under various attacks. To address these issues, the authors propose RingID, a novel multi-channel heterogeneous watermarking approach that combines the advantages of different watermark types. RingID introduces several enhancements, including discretization, lossless imprinting, and spatial shifting, which significantly improve the method's ability to identify multiple keys. RingID achieves substantial improvements in both verification and identification tasks, with verification AUC improving from 0.975 to 0.995 and identification accuracy rising from 0.07 to 0.82. The method is robust to various attacks, including rotation and cropping/scale, and maintains high generation quality. The paper also discusses the design of RingID, including its ability to handle rotation, improve distinguishability through discretization, and increase capacity by using multiple channels. The results show that RingID outperforms Tree-Ring in multi-key identification, demonstrating its effectiveness in enhancing watermark identification capabilities.