As renewable energy sources are becoming more widely integrated into the modern power system, the uncertainties within this system are becoming increasingly prominent. It is crucial to accurately describe the uncertainties in renewable energy output for the effective planning, scheduling, and control of power systems. For this purpose, the aim of this paper is to introduce a method for generating short-term output scenarios for renewable energy sources based on an improved Wasserstein Generative Adversarial Nets-Gradient Penalty. First, a Deep Neural Network structure inspired by the Transformer algorithm is developed to capture the temporal characteristics of renewable energy outputs. Then, combined with the advantage of the data generation of the Wasserstein Generative Adversarial Nets-Gradient Penalty, the Transformer–Wasserstein Generative Adversarial Nets-Gradient Penalty is proposed to generate short-term renewable energy output scenarios. Finally, experimental validation is conducted on open-source wind and photovoltaic datasets from the U.S. National Renewable Energy Laboratory, where the performance of the proposed model in generating renewable energy output scenarios across various aspects (i.e., individual sample representation, expectation and variance, probability density function, cumulative distribution function, power spectral density, autocorrelation coefficient, and pinball loss) is assessed. The results show that our method outperforms the Wasserstein Generative Adversarial Nets-Gradient Penalty, Variational Autoencoder, Copula function, and Latin Hypercube Sampling models in the abovementioned evaluation indicators, providing a more precise probability distribution representation of realistic short-term renewable energy outputs.