SuperAPI AI Directory
Partner with us中文AI newsFavorites
All toolsAI officeAuto-Slides

Auto-Slides

Auto-Slides is an automatic academic presentation generation system proposed by institutions such as the AGI Lab at Westlake University (the official implementation of ICME 2026). Based on large language models and principles of cognitive science, it automatically converts academic research paper PDFs into well-structured, teaching-oriented presentation slides. The system adopts a multi-agent collaboration framework, including PDF parsing, presentation planning, content verification, and automatic repair, supports multimodal output such as images, tables, and code highlighting, and provides natural language interactive customization, multiple Beamer themes, bilingual Chinese-English support, and optional speech script generation. The project is open source under the MIT License and requires a locally installed LaTeX environment and configuration of an OpenAI API key.

Open website

About this tool

Auto-Slides is an intelligent system that automatically converts academic research papers into presentation slides. It was jointly developed by researchers from the AGI Lab at Westlake University and the University of California, Merced, and is the official implementation of the ICME 2026 paper. Its goal is to transform paper content into presentations with a clear structure that conforms to the cognitive principles of teaching, and to support interactive customization.

I. Purpose Auto-Slides is aimed at academic scenarios and can automatically generate structured presentation slides from research paper PDFs. It is suitable for occasions such as academic conference presentations, classroom lectures, and group meeting reports where paper content needs to be transformed into presentation materials. Based on large language models and principles of cognitive science, the system generates presentation-oriented narrative structures to enhance learning and understanding.

II. Main Capabilities

  1. Intelligent PDF processing: Uses OCR and layout analysis technology to automatically extract text, images, tables, and structural information from papers.
  2. Multi-agent framework: Multiple specialized agents are respectively responsible for content extraction, presentation planning, verification, and repair.
  3. Interactive customization: Supports real-time modification and refinement through natural language dialogue.
  4. Teaching optimization: Generates presentation-oriented narrative content to improve learning and understanding.
  5. Multimodal output: Slides include reasonable image placement, table formatting, and code syntax highlighting.
  6. Multi-theme support: Supports multiple Beamer themes (such as Madrid, Berlin, Singapore, etc.).
  7. Bilingual support: Can process English and Chinese papers.
  8. Speech script generation: Optionally generates accompanying speech scripts.

III. How to Use Environment requirements: A LaTeX environment (TeX Live or MiKTeX) is required, as well as more than 8GB of memory (for the marker-pdf model). Installation steps:

  1. Clone the repository;
  2. Create a virtual environment (recommended);
  3. Install dependencies;
  4. Download the required models: Download the marker-pdf model for PDF processing; the model will be downloaded to the models/ directory (about 2GB);
  5. Configure the API key: You can use the provided template file, rename it, and fill in the API key; you can also create a .env file from scratch, create it in the project root directory, and set OPENAI_API_KEY. Basic usage: Convert a research paper into presentation slides. Advanced options:
  • Interactive revision is enabled by default after generating slides and can be disabled with --no-interactive-revise;
  • Existing presentations can be modified based on feedbackDraft. Command-line arguments: Required argument: pdf_path (path to the input PDF file). Optional arguments include: --output-dir/-o (output directory, default output), --language/-l (output language zh or en, default en), --model/-m (language model, default gpt-4o), --theme (Beamer theme, default Madrid), --verbose/-v (show detailed logs), --disable-llm-enhancement (use only basic PDF parsing), --skip-compilation/-s (only generate TEX, skip PDF compilation), --max-retries/-r (maximum number of compilation retries, default 5), --interactive/-i (enable interactive mode for presentation plan optimization), --no-interactive-revise (disable interactive revision after slide generation, enabled by default), --enable-verification (enable content verification, enabled by default), --enable-auto-repair (enable automatic content repair, enabled by default), --disable-verification (disable verification and repair, fast mode), --enable-speech (generate accompanying speech script), --speech-duration (target speech duration, in minutes, default 15), --speech-style (speech style: academic_conference, classroom, industry_presentation, public_talk), --revise/-R (enable revision mode), --original-plan (path to the original presentation plan JSON), --previous-tex (path to the previous TEX file), --feedback (user feedback for modification).

IV. System Architecture Auto-Slides uses a multi-agent framework and includes the following components:

  • PDF Parser: uses marker-pdf and OCR technology to extract content;
  • Presentation Planner: generates a structured presentation plan based on cognitive science principles;
  • Verification Agent: ensures content coverage and accuracy through automatic verification;
  • Repair Agent: automatically fixes detected issues and improves content completeness;
  • TEX Generator: generates high-quality LaTeX Beamer code with standardized formatting;
  • Interactive Editor: enables real-time customization through natural language dialogue. The system goes through multiple stages from content extraction to final presentation generation, with built-in quality assurance and user interaction capabilities.

V. Supported Themes and Languages Themes: Madrid (default), Berlin, Singapore, and more Beamer themes. Languages: English (full support, optimized prompts), Chinese (full support, Chinese-specific processing).

VI. Common Issues and Performance Recommendations

  • Images not extracted or missing: confirm that the marker-pdf model has been downloaded (python down_model.py), and check whether images exist in output/images/<session_id>/.
  • Invalid API key: confirm that OPENAI_API_KEY is set in .env, and check the OpenAI account balance.
  • LaTeX compilation failure: confirm that LaTeX is installed and pdflatex is in PATH; the system automatically uses -shell-escape for code highlighting.
  • Memory issues with large PDFs: the marker-pdf model requires about 8GB of memory; consider using --disable-llm-enhancement for basic processing. Performance recommendations: use --disable-verification to speed up processing; use --skip-compilation to only generate TEX; use --no-interactive-revise to skip interactive features.

VII. Other Information The project uses the MIT License. Citation information: paper arXiv:2509.11062, project page auto-slides.github.io, for issue feedback see GitHub Issues, contact email yangyuheng@westlake.edu.cn. Acknowledgments to marker-pdf, OpenAI, LangChain, and Beamer.

Related tools

Browse category