Generative Adversarial Networks (GANs) are highly successful in areas such as image generation. However, their efficacy in generating time series data, specifically for cloud workload applications, is not well-established. Several GAN architectures are proposed for time series generation, but there is a lack of comprehensive study on performance of these models in cloud workload generation domain. Additionally, prior research has not thoroughly explored the performance of models in relation to dataset attributes, including the length of the data, its seasonality and stationarity. This research addresses these gaps. I conduct a comparative study of four GAN architectures, including TimeGAN, RGAN, TTS-GAN, and V-GAN, using three real-world datasets. The goal is to develop a framework for selecting the best GAN model for cloud workload data generation. I introduce a method to preprocess and characterize the datasets based on existing statistical measures. %, considering varying attributes. To compare the data generated by the models, both qualitatively and quantitatively, I employ them on datasets with diverse characteristics to synthesize data. The synthesized data is then used as input for a microservice application. Response times are measured for both real and synthetic data for comparison. The findings reveal the capabilities and limitations of each model, with regards to input data characteristics. TimeGAN and TTS-GAN are top performing models across various settings. TimeGAN excels at capturing short term temporal dynamics, while TTS-GAN outperforms in capturing long term dependencies. The transformer-based architecture employed in TTS-GAN makes it adept for handling seasonal data across both short and long sequence lengths. Conversely, TimeGAN demonstrates superior performance in capturing seasonal data over shorter periods. The empirical evaluation on the microservice application further confirms the efficacy and applicability of the proposed framework in a realistic testbed setting. This study serves as an empirical guide for practitioners and researchers to choose the most appropriate GAN based on the unique characteristics of their data.