Yuheng Yang¹    Wenjia Jiang¹    Yang Wang¹    Yiwei Wang²    Chi Zhang^1*
1AGI Lab, Westlake University  ²University of California at Merced
[email protected]
     

Auto-Slides is an intelligent system that automatically converts academic research papers into well-structured, pedagogically optimized presentation slides. Built on large language models and cognitive science principles, it creates multimodal presentations with interactive customization capabilities.

• Intelligent PDF Processing: Automatically extracts text, figures, tables, and structure from research papers using advanced OCR and layout analysis
• Multi-Agent Framework: Employs specialized agents for content extraction, presentation planning, verification, and repair
• Interactive Customization: Supports real-time refinement through natural language dialogue
• Pedagogical Optimization: Creates presentation-oriented narratives that enhance learning and comprehension
• Multimodal Output: Generates slides with proper figure placement, table formatting, and code syntax highlighting
• Multiple Themes: Supports various Beamer themes (Madrid, Berlin, Singapore, etc.)
• Bilingual Support: Works with both English and Chinese papers
• Speech Generation: Optional accompanying speech script generation

• Python 3.8+
• LaTeX environment (TeX Live or MiKTeX)
• OpenAI API key
• 8GB+ RAM (for marker-pdf model)

git clone https://github.com/your-username/auto-slides.git
cd auto-slides

python -m venv venv
source venv/bin/activate  # Linux/Mac
# or
venv\Scripts\activate  # Windows

pip install -r requirements.txt

Important: Download the marker-pdf model for PDF processing:

python down_model.py

This will download the marker-pdf model to the models/ directory (~2GB).

Option 1: Use the provided template (Recommended)

Rename the provided template file and add your API key:

# Rename the template file
mv ".env copy" .env

# Edit the .env file and add your OpenAI API key
nano .env  # or use your preferred editor

Option 2: Create from scratch

Create a .env file in the project root:

# OpenAI API Configuration (Required)
OPENAI_API_KEY=your_openai_api_key_here

# Optional: LangSmith for monitoring (uncomment if needed)
# LANGCHAIN_TRACING_V2="true"
# LANGCHAIN_ENDPOINT="https://api.smith.langchain.com"
# LANGCHAIN_API_KEY="your_langsmith_api_key_here"
# LANGCHAIN_PROJECT="auto-slides"

Note: The .env copy file contains a template with all available configuration options. Simply rename it to .env and update the OPENAI_API_KEY value with your actual API key.

Convert a research paper to presentation slides:

python main.py path/to/your/paper.pdf

python main.py path/to/your/paper.pdf \
    --language en \
    --model gpt-4o \
    --theme Madrid \
    --output-dir output \
    --verbose

By default, Auto-Slides enables interactive revision after generating slides. To disable this feature:

python main.py path/to/your/paper.pdf --no-interactive-revise

Modify existing presentations with feedback:

python main.py --revise \
    --original-plan=path/to/plan.json \
    --previous-tex=path/to/output.tex \
    --feedback="Please make the title more prominent"

• pdf_path: Path to the input PDF file

• --output-dir, -o: Output directory (default: output)
• --language, -l: Output language (zh or en, default: en)
• --model, -m: Language model to use (default: gpt-4o)
• --theme: Beamer theme (default: Madrid)
• --verbose, -v: Show verbose logs

• --disable-llm-enhancement: Use basic PDF parsing only
• --skip-compilation, -s: Generate TEX only, skip PDF compilation
• --max-retries, -r: Maximum retries for compilation (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 (default: enabled)
• --enable-auto-repair: Enable automatic content repair (default: enabled)
• --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 original presentation plan JSON
• --previous-tex: Path to previous TEX file
• --feedback: User feedback for modifications

The system generates organized output in the specified directory:

output/
├── raw/<session_id>/          # Extracted PDF content
├── plan/<session_id>/         # Presentation plans (JSON)
├── tex/<session_id>/          # Generated LaTeX files and PDFs
├── images/<session_id>/       # Extracted figures and tables
├── verification/<session_id>/ # Verification reports
├── repair/<session_id>/       # Auto-repair results
└── speech/<session_id>/       # Generated speech scripts

Auto-Slides employs a sophisticated multi-agent framework with the following components:

1. PDF Parser: Extracts content using marker-pdf and OCR technologies
2. Presentation Planner: Generates structured presentation plans based on cognitive science principles
3. Verification Agent: Ensures content coverage and accuracy through automated validation
4. Repair Agent: Automatically fixes identified issues and improves content completeness
5. TEX Generator: Creates high-quality LaTeX Beamer code with proper formatting
6. Interactive Editor: Enables real-time customization through natural language dialogue

The system processes research papers through multiple stages, from initial content extraction to final presentation generation, with built-in quality assurance and user interaction capabilities.

• Madrid (default)
• Berlin
• Singapore
• Warsaw
• Copenhagen
• And more Beamer themes

• English: Full support with optimized prompts
• Chinese: Full support with Chinese-specific processing

Q: Images not extracted or missing?

• Ensure the marker-pdf model is downloaded: python down_model.py
• Check that images exist in output/images/<session_id>/

Q: API key not working?

• Verify OPENAI_API_KEY is set in your .env file
• Check your OpenAI account has sufficient credits

Q: LaTeX compilation fails?

• Ensure LaTeX is installed and pdflatex is in your PATH
• The system automatically uses -shell-escape for code highlighting

Q: Memory issues with large PDFs?

• The marker-pdf model requires ~8GB RAM
• Consider using --disable-llm-enhancement for basic processing

• Use --disable-verification for faster processing
• Enable --skip-compilation to generate TEX only
• Use --no-interactive-revise to skip interactive features

If you use Auto-Slides in your research, please cite our paper:

@article{yang2025autoslides,
  title={Auto-Slides: Automatic Academic Presentation Generation with Multi-Agent Collaboration},
  author={Yang, Yuheng and Jiang, Wenjia and Wang, Yang and Wang, Yiwei and Zhang, Chi},
  journal={arXiv preprint arXiv:2509.11062},
  year={2025},
  note={AGI Lab, Westlake University; University of California at Merced; Corresponding author: Chi Zhang},
  url={https://auto-slides.github.io/},
  eprint={2509.11062},
  archivePrefix={arXiv},
  primaryClass={cs.AI}
}

This project is licensed under the MIT License - see the LICENSE file for details.

Contributions are welcome! Please feel free to submit a Pull Request.

• Project Page: auto-slides.github.io
• Paper: arXiv:2509.11062
• Issues: GitHub Issues
• Email: [email protected]

• marker-pdf for PDF processing
• OpenAI for language models
• LangChain for LLM orchestration
• Beamer for presentation framework