The paper introduces a novel approach to generating weather forecast ensembles using deep generative diffusion models, which aims to reduce computational costs associated with traditional physics-based ensemble forecasting. The authors propose the Scalable Ensemble Envelope Diffusion Sampler (SEEDS), a method that can generate large ensembles of weather forecasts with minimal computational resources. SEEDS is trained on historical data from operational numerical weather prediction (NWP) models, specifically the Global Ensemble Forecast System (GEFS) version 12, and the ERA5 reanalysis dataset from the European Centre for Medium-Range Weather Forecasts (ECMWF). The generated ensembles are compared to ground-truth ensembles from operational systems and show similar statistical properties and predictive skills, while being more computationally efficient. The method is particularly useful for forecasting rare and extreme weather events, as it can provide more accurate likelihoods and better characterizes spatial patterns associated with these events. The paper also discusses the reliability and predictive skill of the generated ensembles, showing improved reliability and better skill in predicting extreme events compared to physics-based ensembles. The authors conclude that SEEDS offers a promising approach to enhancing the efficiency and accuracy of weather forecasting, with potential applications in climate risk assessment.The paper introduces a novel approach to generating weather forecast ensembles using deep generative diffusion models, which aims to reduce computational costs associated with traditional physics-based ensemble forecasting. The authors propose the Scalable Ensemble Envelope Diffusion Sampler (SEEDS), a method that can generate large ensembles of weather forecasts with minimal computational resources. SEEDS is trained on historical data from operational numerical weather prediction (NWP) models, specifically the Global Ensemble Forecast System (GEFS) version 12, and the ERA5 reanalysis dataset from the European Centre for Medium-Range Weather Forecasts (ECMWF). The generated ensembles are compared to ground-truth ensembles from operational systems and show similar statistical properties and predictive skills, while being more computationally efficient. The method is particularly useful for forecasting rare and extreme weather events, as it can provide more accurate likelihoods and better characterizes spatial patterns associated with these events. The paper also discusses the reliability and predictive skill of the generated ensembles, showing improved reliability and better skill in predicting extreme events compared to physics-based ensembles. The authors conclude that SEEDS offers a promising approach to enhancing the efficiency and accuracy of weather forecasting, with potential applications in climate risk assessment.