Our proposed approach aims to automate the transformation of textual content into visually engaging and multilingual videos through a systematic workflow. Users input article content into our website, and if the content exceeds a specified length, it undergoes summarization using the 'pegasus-cnn_dailymail' technique. The summarized text serves as input to our LLaMA model, a custom fine-tuned language model based on GPT-3, which generates a video script by analyzing the content's key messages and structuring them logically. The generated script undergoes user review and editing for refinement. Subsequently, the script is translated into 13 official languages using Google Translate, ensuring broader accessibility. Simultaneously, the LLaMA model generates keywords to search the web for relevant images, and the user curates a selection that aligns with the narration. Microsoft's text-to-speech engine synthesizes voiceovers in all 13 languages. The approved images and synthesized voice overs are combined using the MoviePy library to create a cohesive video. The video is then rendered in each of the 13 languages. The entire process, from script generation to multilingual video rendering, takes approximately 7-8 minutes. Our methodologies incorporate thoughtful considerations for text length, narration speed, and content relevance to ensure a balanced and engaging video output. The system's design aims to automate manual tasks and streamline the video creation process. The paper details the complete workflow and technologies employed, emphasizing the effectiveness of the 'pegasus-cnn_dailymail' summarization technique and the LLaMA model in generating high-quality scripts for multilingual video production.

Key Contributions

• Automated Text-to-Video Workflow
• Multilingual Support in 13 Languages
• Custom Fine-Tuned LLaMA Model
• Integration of Text Summarization
• User-Centric Content Curation