University tool turns research papers into TikTok clips

PaperTok turns – PaperTok, a new platform from the University of Washington’s Prosocial Computing Group, turns dense academic research PDFs into short 45-second science videos for general audiences using Google’s Gemini. Designed after researchers noticed the public using gene

For a few months, the same pattern kept showing up online: short science videos created with generative AI, posted by people who weren’t scientists. The videos were easy to make, quick to share—and, to researchers at the University of Washington, worryingly prone to mistakes.

That tension is what pushed the University of Washington’s Prosocial Computing Group to build PaperTok, a platform meant to help the public engage with research without the accuracy gap that comes when AI content is created by those unfamiliar with the source material.

PaperTok is designed to take a scientific paper and turn it into a short video—about 45 seconds—using Gemini. Users upload an academic PDF. the system analyzes the content. and PaperTok generates a video script and clips that can be edited before the final video is created. At the moment, PaperTok is only accessible with a paid Google Gemini subscription.

Senior author Gary Hsieh. a UW professor in human centered design and engineering. said in a release that the problem wasn’t just the existence of AI-generated summaries—it was that most people don’t read research papers in the first place. He also described the difficulty he personally faces when reading papers outside his field. saying he “still have challenges reading papers in fields I’m not familiar with.” The aim. he explained. was to turn papers into a format laypeople would want to engage with. and to study how they engage with it.

The team brought the work to a major audience in April, presenting their research at the Association for Computing Machinery Conference on Human Factors in Computing Systems in Barcelona.

The tool’s workflow is built to keep users involved at key points rather than treating the output as final. The process starts on the PaperTok site: users upload a PDF and select a script. The system automatically analyzes the paper and produces four possible options for the video’s hook—its lead—so the user can choose the entry point. PaperTok then writes a script that users can edit, refining both the wording and the voiceover tone in an editor.

From there, the script drives the visuals. Once the script is complete, users generate video scenes in a storyboard format, with clips broken into scenes like a movie outline. Users can refine and edit the scripts alongside the matching video clips as the storyboard comes together.

In the last stage, PaperTok handles the production wrap: it automatically generates screen credits from the paper’s original authors, and users can add their own byline. The scenes are then merged into a shareable short-form video ready for posting.

The tight loop—from PDF to hook to edited script to scene-by-scene clips—directly answers the concern that originally caught UW researchers’ attention: AI science videos made by non-specialists are gaining traction online. and PaperTok is built to route that trend through the actual authorship of the paper. with user control to reduce the risk of a nonscientist’s misstep.

PaperTok University of Washington Prosocial Computing Group Gemini generative AI academic research TikTok science videos human centered design and engineering science communication