Automated video generation from user text prompts has recently achieved remarkable progress in both the quality and duration of generated content. Text-to-video generation models transform minimal textual instructions into animated videos with rich scenes, dynamic actions, and coherent narratives. This transformation is driven by evolving methodological approaches, with large language models (LLMs) playing a central role due to their exceptional ability to perform reasoning and generate structured text. LLMs effectively convert simple textual instructions into detailed scenario descriptions and visual sequences, making them a core component of contemporary video generation pipelines. This literature review examines recent advances in leveraging LLMs within video generation pipelines, focusing on how they are integrated to guide content synthesis and on the technical challenges encountered. Key findings indicate that LLMs enhance semantic coherence, narrative richness, and content diversity in generated videos, while significant challenges remain, including high computational costs, hallucinations, and maintaining spatiotemporal consistency.