Scholarlink

This study introduces Slide2Vid, a novel methodology for converting static slide presentations into dynamic, contextually rich video content using large language models. Addressing the limitations of existing methods, Slide2Vid encompasses three key stages: input data preprocessing, initial context generation, and an innovative iterative refinement process to enhance narrative coherence and quality, which is known as Sequential Contextual Refinement through Swapping (SCRS). By leveraging the strengths of large language models and the strategic sequencing of these stages, Slide2Vid offers a robust solution for creating engaging video content from slide presentations. Our evaluation, with various focused metrics, demonstrates significant improvements in semantic coherence and overall narrative quality, with the final SCRS stage producing the most polished outputs. These findings highlight Slide2Vid’s potential to bridge the gap between static and dynamic media, offering a comprehensive slide-to-video conversion method. While the study shows promising potential, it also acknowledges limitations, such as reliance on computationally intensive models and a small dataset, suggesting future research directions to validate and enhance the methodology across diverse content types and presentation styles. Here is our Colab Notebook for the project:

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Slide2Vid: Dynamic Video Generation from Static Presentations Through Sequential Contextual Refinement

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Tri-Quang Pham,
  • Xuan-Vinh Vu,
  • Thien-Anh Nguyen,
  • Cong-Thien Pham,
  • Thanh-Tho Quan

摘要

This study introduces Slide2Vid, a novel methodology for converting static slide presentations into dynamic, contextually rich video content using large language models. Addressing the limitations of existing methods, Slide2Vid encompasses three key stages: input data preprocessing, initial context generation, and an innovative iterative refinement process to enhance narrative coherence and quality, which is known as Sequential Contextual Refinement through Swapping (SCRS). By leveraging the strengths of large language models and the strategic sequencing of these stages, Slide2Vid offers a robust solution for creating engaging video content from slide presentations. Our evaluation, with various focused metrics, demonstrates significant improvements in semantic coherence and overall narrative quality, with the final SCRS stage producing the most polished outputs. These findings highlight Slide2Vid’s potential to bridge the gap between static and dynamic media, offering a comprehensive slide-to-video conversion method. While the study shows promising potential, it also acknowledges limitations, such as reliance on computationally intensive models and a small dataset, suggesting future research directions to validate and enhance the methodology across diverse content types and presentation styles. Here is our Colab Notebook for the project: