Gaussian Shading is a performance-lossless image watermarking technique for diffusion models that enables copyright protection and tracing of malicious content without altering model performance or requiring additional training. The method embeds watermarks in latent representations following a standard Gaussian distribution, making them indistinguishable from non-watermarked images. This approach ensures that the watermark is seamlessly integrated into the generation process without affecting the model's performance. The watermark is robust to lossy processing and erasure attempts, and can be extracted using Denoising Diffusion Implicit Models (DDIM) inversion and inverse sampling. Gaussian Shading outperforms existing methods in terms of robustness and achieves a high-capacity watermark of 256 bits. The method is theoretically proven to be performance-lossless and can be seamlessly integrated into the generation process as a plug-and-play module. Experiments on multiple versions of Stable Diffusion demonstrate that Gaussian Shading maintains high performance and robustness under various attacks, including compression and inversion attacks. The method is also effective in both detection and traceability scenarios, with high accuracy in identifying and tracing watermarked images. Despite its advantages, Gaussian Shading has limitations, including reliance on DDIM inversion and the need for proper key management. Future work includes improving inversion methods and expanding the range of sampling methods to counter forgery attacks.Gaussian Shading is a performance-lossless image watermarking technique for diffusion models that enables copyright protection and tracing of malicious content without altering model performance or requiring additional training. The method embeds watermarks in latent representations following a standard Gaussian distribution, making them indistinguishable from non-watermarked images. This approach ensures that the watermark is seamlessly integrated into the generation process without affecting the model's performance. The watermark is robust to lossy processing and erasure attempts, and can be extracted using Denoising Diffusion Implicit Models (DDIM) inversion and inverse sampling. Gaussian Shading outperforms existing methods in terms of robustness and achieves a high-capacity watermark of 256 bits. The method is theoretically proven to be performance-lossless and can be seamlessly integrated into the generation process as a plug-and-play module. Experiments on multiple versions of Stable Diffusion demonstrate that Gaussian Shading maintains high performance and robustness under various attacks, including compression and inversion attacks. The method is also effective in both detection and traceability scenarios, with high accuracy in identifying and tracing watermarked images. Despite its advantages, Gaussian Shading has limitations, including reliance on DDIM inversion and the need for proper key management. Future work includes improving inversion methods and expanding the range of sampling methods to counter forgery attacks.