Architectural Framework for Multi-Modal Video Generation via Fine-Tuned Stable Diffusion Models

In recent years, large language models (LLMs) like ChatGPT have driven advancements in artificial intelligence (AI) across various domains, pushing the boundaries of multi-modal applications. The emergence of Stable Diffusion XL in April 2023 brought image generation into the mainstream. The generation of videos from textual input poses a significant computational challenge within computer science. Nonetheless, recent advancements in text-to-video artificial intelligence (AI) technologies have showcased notable progress within this domain. [1] Following the rapid development of LLMs, OpenAI's release of the Sora promotional video in February 2024 marked a significant milestone in video generation. In this paper, we explore a novel approach to generating coherent videos by fine-tuning Stable Diffusion, enabling the production of temporally consistent video frames. Our architecture integrates LLM-based text expansion, multi-frame generation using Stable Diffusion, and final video synthesis. Experimental results demonstrate that our method significantly improves the temporal coherence of generated videos.