Abstract: Renewable energy represented by photovoltaic and wind power has characteristics of strong intermittence and fluctuation, which brings great challenges to the operation and optimization of distribution networks. To investigate the uncertainties of photovoltaic and wind power output, a stochastic scenario generation method for renewable energy based on the pixel convolutional generative network (PCGN) is proposed. The joint distribution of photovoltaic and wind power generation curves is transformed into the product of multiple one-dimensional distributions by the chain rule, and the previously generated output power is adopted to predict the next power value, thus achieving point-by-point generation of each power point in power generation curves of renewable energies. According to the high-dimensional data characteristics of photovoltaic and wind power generation curves, network structures applicable to the stochastic scenario generation of the renewable energy is designed, and the effectiveness and adaptability of the proposed method are verified by the real operation data. Simulation results show that the proposed PCGN can not only accurately capture the shape characteristics, volatility, probability distribution, and spatio-temporal correlation of photovoltaic and wind power curves, but also has strong adaptability, which can be used for stochastic scenario generation tasks for different renewable energies by fine-tuning the structure and parameters of generative networks.