DiffuseTrace is a novel watermarking scheme for latent diffusion models (LDMs) that embeds invisible watermarks semantically in all generated images for future detection. Unlike existing methods that rely on post-processing and require model retraining for watermark message changes, DiffuseTrace embeds watermarks into the initial latent variables through an encoder-decoder model, ensuring semantic consistency and image quality. The watermark is integrated into the sampling process and extracted by reversing the diffusion process using the decoder. DiffuseTrace does not require fine-tuning of the diffusion model components and can be used as a plug-in in arbitrary diffusion models. The scheme is robust against attacks based on variational autoencoders and diffusion models, and it maintains high accuracy and flexibility. The watermark is embedded into the image space semantically without compromising image quality. The encoder-decoder can be utilized as a plug-in in arbitrary diffusion models. The scheme is validated through experiments, demonstrating its effectiveness and flexibility. DiffuseTrace holds an unprecedented advantage in combating the latest attacks based on variational autoencoders and diffusion models. The primary contributions of this work are: (1) DiffuseTrace is the first scheme that embeds robust multi-bit watermarks in diffusion models without relying on the trade-off between image quality and watermark robustness. (2) DiffuseTrace exhibits significant performance in common image processing and robustness against attacks based on variational autoencoders and diffusion models. (3) The proposed universal watermark module for latent diffusion models can be seamlessly integrated across different versions of diffusion models. The watermark message of DiffuseTrace can be flexibly modified without being affected by model fine-tuning or model update iterations. The scheme is validated through experiments, demonstrating its effectiveness and flexibility. DiffuseTrace holds an unprecedented advantage in combating the latest attacks based on variational autoencoders and diffusion models.DiffuseTrace is a novel watermarking scheme for latent diffusion models (LDMs) that embeds invisible watermarks semantically in all generated images for future detection. Unlike existing methods that rely on post-processing and require model retraining for watermark message changes, DiffuseTrace embeds watermarks into the initial latent variables through an encoder-decoder model, ensuring semantic consistency and image quality. The watermark is integrated into the sampling process and extracted by reversing the diffusion process using the decoder. DiffuseTrace does not require fine-tuning of the diffusion model components and can be used as a plug-in in arbitrary diffusion models. The scheme is robust against attacks based on variational autoencoders and diffusion models, and it maintains high accuracy and flexibility. The watermark is embedded into the image space semantically without compromising image quality. The encoder-decoder can be utilized as a plug-in in arbitrary diffusion models. The scheme is validated through experiments, demonstrating its effectiveness and flexibility. DiffuseTrace holds an unprecedented advantage in combating the latest attacks based on variational autoencoders and diffusion models. The primary contributions of this work are: (1) DiffuseTrace is the first scheme that embeds robust multi-bit watermarks in diffusion models without relying on the trade-off between image quality and watermark robustness. (2) DiffuseTrace exhibits significant performance in common image processing and robustness against attacks based on variational autoencoders and diffusion models. (3) The proposed universal watermark module for latent diffusion models can be seamlessly integrated across different versions of diffusion models. The watermark message of DiffuseTrace can be flexibly modified without being affected by model fine-tuning or model update iterations. The scheme is validated through experiments, demonstrating its effectiveness and flexibility. DiffuseTrace holds an unprecedented advantage in combating the latest attacks based on variational autoencoders and diffusion models.