Scholarlink

In the recent development of Artificial Intelligence (AI) and Machine Learning (ML), machines have started to process and interpret both visual and textual data with high precision. Such developments are deployed in domains like healthcare, autonomous vehicles, NLP, and creative industries. The domain of this article is the creative industry, where the focus is on developing a robust framework for transforming images into coherent narratives. This is implemented using Vision Transformers (ViT) and the GPT-2 language model. It combines Vision Transformers (ViT) for feature extraction and GPT-2 for semantically rich caption generation to overcome object hallucinations and misalignment. It builds knowledge graphs with NetworkX for organizing entity relationships, enabling multi-image integration and contextual analysis. The Cohere API then generates stories from the graphs, providing scalable, flexible storytelling with real-time updates and multi-domain potential. This system fills the critical gaps in visual content interpretation. It will have transformative applications in areas such as data-driven journalism, educational tools, social media content creation, and automated storytelling. In addressing the limitations of static image captioning and advancing dynamic storytelling, this research shows how AI can gain deeper insights and meaningful narratives from visual data.

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

From Pixels to Stories: A Scalable AI Framework for Dynamic Image Narratives Using Vision Transformers and GPT-2

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • K. R. Sayanthika,
  • R. K. Adithya Nambiar,
  • Alap S. Suresh,
  • B. R. Gokul Krishna,
  • Lekshmi S. Nair

摘要

In the recent development of Artificial Intelligence (AI) and Machine Learning (ML), machines have started to process and interpret both visual and textual data with high precision. Such developments are deployed in domains like healthcare, autonomous vehicles, NLP, and creative industries. The domain of this article is the creative industry, where the focus is on developing a robust framework for transforming images into coherent narratives. This is implemented using Vision Transformers (ViT) and the GPT-2 language model. It combines Vision Transformers (ViT) for feature extraction and GPT-2 for semantically rich caption generation to overcome object hallucinations and misalignment. It builds knowledge graphs with NetworkX for organizing entity relationships, enabling multi-image integration and contextual analysis. The Cohere API then generates stories from the graphs, providing scalable, flexible storytelling with real-time updates and multi-domain potential. This system fills the critical gaps in visual content interpretation. It will have transformative applications in areas such as data-driven journalism, educational tools, social media content creation, and automated storytelling. In addressing the limitations of static image captioning and advancing dynamic storytelling, this research shows how AI can gain deeper insights and meaningful narratives from visual data.