Multidimensional Text-to-Video Generation: Integrating Sentiment, Pragmatics, and Semantics

Abstract

In this study, we have developed a new framework for creating videos automatically. Unlike diffusion models, our work focuses on assembly and composition of existing media rather than generation of new content with high overhead.   The approach followed in this research work deconstructs the input script into sentences and examines each textual input segment using sentiment, pragmatic, and semantic analysis. The query engine receives entities found in the love letter content—the use case investigated—and gathers pertinent video clips. Our method aligns the segments for meaningful links, creating a seamless video composition that mirrors the written story. The work gives better qualitative outcomes as it incorporates sentiment and pragmatic analysis. We used a panel of five judges to assess the quality of the generated automated movies, and computed an intraclass correlation coefficient to ascertain inter-rater agreement. In terms of the produced videos & quality, we got encouraging results. The successful outcome shows that this is an efficient method (very low overhead as compared to diffusion models)  for automating contextual video synthesis based on limited text input, as further confirmed by the panel consensus and our cohesion-focused methodology.