University of Washington Researchers Launch PaperTok AI Tool to Turn Research Papers Into Videos

University of Washington Researchers Build AI-Powered Tool to Fight Misinformation

A team from the University of Washington's Prosocial Computing Group has launched PaperTok, an AI tool designed to help scientists turn research papers into videos and reclaim science communication from non-experts spreading misinformation. The initiative emerged after doctoral students noticed a troubling trend: members of the general public were using generative AI to create short science videos without scientific expertise, accelerating the spread of what researchers call AI slop 1

1

. "The alternative is that science is being talked about without scientists," said co-lead author Meziah Ruby Cristobal, a UW doctoral student in human centered design and engineering 3

3

.

How PaperTok Transforms Academic Papers Into Short Engaging Videos

The platform streamlines science communication by converting dense, jargon-heavy publications into 45-second clips suitable for TikTok videos, YouTube Shorts, and Instagram Reels. Researchers upload academic papers to PaperTok, which uses Google Gemini to analyze the content and identify attention-grabbing hooks and relevant takeaways for general audiences 1

1

. The AI tool generates four hook options—such as "Ever get overwhelmed reading a dense academic paper?"—before creating a complete script with an opening scene and narrative arc 3

3

. Each video closes with a byline referencing the paper, authors' names, and journal to establish credibility. Currently, PaperTok requires users to have a paid Google Gemini subscription and a Gemini API key, as video generation is computationally expensive 1

1

.

Keeping Humans in the Loop to Combat AI-Generated Misinformation

Unlike other PDF-to-video generators, PaperTok was intentionally designed to maintain human oversight at every stage. The multi-step process requires researcher approval at each phase, allowing users to edit output down to individual words 1

1

. During the storyboarding phase, scripts break into scenes that users can continuously refine along with matching video clips 3

3

. "If anything, we hope that PaperTok highlights how important people are in science communication," Cristobal emphasized, addressing concerns about misinformation risks from fully automated systems 3

3

.

Research Validation and User Feedback Shape Development

Cristobal presented research on PaperTok at the ACM CHI Conference on Human Factors in Computing Systems in Barcelona, co-leading the study with fellow doctoral student Donghoon Shin under senior author Gary Hsieh, a UW professor in human centered design and engineering 1

1

. A team of eight built the platform last summer, starting with interviews with eight science communicators and content producers before gathering user feedback 1

1

. In a user study, 100 online participants and 18 academic participants compared PaperTok videos with those from two other systems, finding PaperTok easier to use and more engaging 3

3

. Many researchers discovered unexpected value in seeing how AI visualizes abstract concepts, treating it as a brainstorming tool 1

1

.

Addressing Concerns and Expanding Accessible Scientific Research

Some users expressed concerns that videos felt "too AI-ish" due to issues like nonsense text, worrying this might diminish their scholarship's credibility if shared publicly 3

3

. The team continues refining the platform, with plans to let researchers incorporate charts and graphics from their papers and customize AI-generated video by drawing on specific scene elements 3

3

. While originally built for human-computer interaction research papers, PaperTok has been tested on physics topics and performed well 1

1

. The team aims to expand across research disciplines, creating videos for social sciences and hard sciences alike. As short-form science content proliferates across platforms, PaperTok offers researchers a way to maintain control over their narratives while adapting to formats that laypeople engage with, directly addressing the challenge of making complex research accessible without sacrificing accuracy.