Diffusion-TS is a novel diffusion-based framework designed to generate high-quality multivariate time series samples. It combines an encoder-decoder transformer with disentangled temporal representations, guided by a decomposition technique that captures the semantic meaning of time series. Unlike existing diffusion-based approaches, Diffusion-TS trains the model to directly reconstruct the sample rather than the noise in each diffusion step, incorporating a Fourier-based loss term. This approach enhances the interpretability and realism of the generated time series. Diffusion-TS is also extendable to conditional generation tasks such as forecasting and imputation without requiring additional model changes. Experiments demonstrate that Diffusion-TS achieves state-of-the-art results on various realistic time series datasets, showing superior performance in terms of diversity, novelty, and interpretability. The model's ability to handle complex time series, including those with seasonal oscillations, is highlighted, along with its effectiveness in challenging settings with limited data.Diffusion-TS is a novel diffusion-based framework designed to generate high-quality multivariate time series samples. It combines an encoder-decoder transformer with disentangled temporal representations, guided by a decomposition technique that captures the semantic meaning of time series. Unlike existing diffusion-based approaches, Diffusion-TS trains the model to directly reconstruct the sample rather than the noise in each diffusion step, incorporating a Fourier-based loss term. This approach enhances the interpretability and realism of the generated time series. Diffusion-TS is also extendable to conditional generation tasks such as forecasting and imputation without requiring additional model changes. Experiments demonstrate that Diffusion-TS achieves state-of-the-art results on various realistic time series datasets, showing superior performance in terms of diversity, novelty, and interpretability. The model's ability to handle complex time series, including those with seasonal oscillations, is highlighted, along with its effectiveness in challenging settings with limited data.