By Gaurab Chhetri - Personal Project
VidXiv - ArXiv Paper to Video Generator
VidXiv automatically converts research papers from ArXiv into engaging narrated videos with scene-by-scene breakdowns, AI-generated scripts, and export-ready MP4s for YouTube or Shorts.
VidXiv is a tool I created to make research more accessible. Academic papers are essential for progress, but they are often dense and intimidating for people outside of research. VidXiv bridges that gap by turning ArXiv papers into short, narrated videos, something that can be shared on YouTube, TikTok, or Instagram Reels.
The goal is simple: help researchers, educators, and science communicators transform complex ideas into engaging visual stories that reach a wider audience.
What It Does
- Fetches papers by ArXiv ID and parses the PDF content
- Generates a video script with AI (Gemini), summarizing each section into digestible narration
- Builds scene-by-scene videos with text overlays, transitions, and optional background music
- Narrates with TTS voices so the videos are watchable without extra editing
- Supports multiple formats (16:9 for YouTube, 9:16 for Shorts/Reels)
Here’s the flow: you paste a paper ID → AI writes a script → VidXiv builds scenes → you download a ready-to-share MP4.
Example Workflow
- Enter a paper ID (e.g.,
2401.06015) into the Streamlit app - AI condenses the paper into a script with scene segments
- Text overlays and narration are generated for each scene
- Figures and diagrams (future feature) can be auto-extracted or manually added
- Export as a polished video ready for YouTube or social platforms
Why I Built It
Researchers work hard to publish, but outside academia, papers rarely get read. I wanted to create a tool that makes knowledge portable, shareable, and accessible in the formats people already consume daily, short videos.
VidXiv is also a learning project for me, blending LLM prompting, PDF parsing, text-to-speech, and video editing into one pipeline. It is both a research communication tool and a technical challenge in AI + media generation.
Impact
VidXiv is a step toward making science more open and approachable. Instead of a 15-page PDF, someone can watch a 2-minute explainer. For me, it is also part of a larger vision of blending AI, research, and storytelling to democratize knowledge.
The AIT Lab at Texas State University 1 is already using this tool to summarize the survey papers. See https://ait-lab.vercel.app/story/survey-tinyml for example.