Paper2Video - An Academic Paper-to-Demo Video Generation Project Launched by the National University of Singapore

What Is Paper2Video

Paper2Video is a project developed by the Show Lab at the National University of Singapore that automatically generates presentation videos from academic papers. Through the PaperTalker multi-agent framework, it transforms papers into complete presentation videos containing slides, subtitles, speech, and a presenter avatar. The framework includes four modules: a slide builder, subtitle builder, cursor builder, and presenter builder, which are responsible for slide generation, subtitle generation, cursor positioning, and presenter video generation, respectively. Paper2Video provides the first high-quality academic presentation video benchmark, including data such as 101 papers and their corresponding author presentation videos and slides. The benchmark also designs four evaluation metrics—Meta Similarity, PresentArena, PresentQuiz, and IP Memory—to measure whether a video accurately conveys the core ideas of a paper, whether it is easy to understand, whether it highlights the authors' contributions, and whether it enhances the impact of the research.

Main Features of Paper2Video

• Automatic Video Generation: Automatically generates presentation videos from academic papers, transforming complex academic content into easily understandable visual and auditory formats.
• Multi-Agent Framework: Through the PaperTalker framework, it integrates multiple modules such as slide generation, subtitle generation, cursor positioning, speech synthesis, and presenter avatar rendering to achieve efficient and high-quality video production.
• High-Quality Benchmark Dataset: Provides a benchmark containing data such as 101 papers and their authors' presentation videos and slides, offering a standard for research and evaluation of academic presentation videos.
• Customized Evaluation Metrics: Designs evaluation metrics such as Meta Similarity, PresentArena, PresentQuiz, and IP Memory to measure the quality and effectiveness of presentation videos from different perspectives.
• Easy-to-Use Tool: Provides complete code and detailed usage guides, making it convenient for researchers and developers to use the tool to generate their own presentation videos.

Technical Principles of Paper2Video

• Slide Generation and Optimization: Extracts content from a paper's LaTeX source files and generates Beamer-format slide drafts. It uses a "tree-search visual selection" method to optimize layouts: the system generates multiple layout candidates and then lets a vision-language model (VLM) judge the best version.
• Subtitle and Cursor Generation: Generates corresponding narration (subtitles) for slides and plans the trajectory of a simulated presenter's mouse cursor movement during the presentation. The cursor movement and speech are precisely aligned in time and space to guide the audience's attention.
• Presenter Generation: Uses a portrait photo of the author and a short voice sample to synthesize a virtual avatar with the author's personal characteristics and lip synchronization with speech through text-to-speech (TTS) and talking-face generation technologies.
• Parallel Processing: Splits the video generation task by slide and processes them in parallel, greatly shortening the total generation time.

Paper2Video Project Address

• Project Official Website：https://showlab.github.io/Paper2Video/
• GitHub Repository：https://github.com/showlab/Paper2Video
• arXiv Technical Paper：https://arxiv.org/pdf/2510.05096

Application Scenarios of Paper2Video

• Academic Conferences: Provides researchers with a tool to quickly generate high-quality presentation videos, saving preparation time and improving presentation effectiveness.
• Online Courses: Help educators transform the content of academic papers into vivid video courses, enhancing teaching interactivity and appeal.
• Social Media Dissemination: Enable academic achievements to be shared on social media in a more accessible video format, expanding research impact.
• Academic Presentations: Make it easy for researchers to quickly produce academic presentation videos for internal briefings or public lectures.
• Research Promotion: Provide research institutions and scholars with a new way to showcase research findings, enhancing research visibility and public awareness.