“Computational story creation” aims to teach computers how to imitate the human imagination. It uses a various attention models in artificial intelligence (AI) and psychology to generate narrative sentences. The challenge is to produce text with a recurring theme and innovative terminology, as well as automatically develop natural language. Due to its strong creative potential, Generative Pretrained Transformer 2 (GPT-2) has often been incorporated into narrative creation models. But it continues to lack variety and provide contradictory tales. Current story creation models incorporate other data, including plots or intelligence, into GPT-2 to direct the generating process. Although the Conditional Variational Autoencoder (CVAE) module makes it possible to increase phrase diversity, the keyword search planning technique is used to support thematic coherence. All these methods concentrate on increasing the quality of new stories. Although in this review we concentrated on the basic design of CVAE and VAE architectures, the attention framework, and the model comparison of CVAE and VAE technologies, future researchers might make use of this work to enhance the implementation of the autoencoder technique in narrative creation to improve results. This model is useful for automated story creation like online gaming, filmmaking, writing novels, and educational applications in the current digital era.