Transform any arXiv papers into slides using LLMs! This tool is useful for quickly grasping the main ideas of research papers. Some examples of generated slides are in the Demo directory.

Python 3.10 or higher is required.

To install:

1. Clone this repository:

   git clone https://github.com/takashiishida/paper2slides.git
   cd paper2slides

2. Install the required Python packages:

   pip install -r requirements.txt

3. Ensure pdflatex is installed and available in your system's PATH. Optionally check if you can compile the sample test.tex by pdflatex test.tex. Check if test.pdf is generated correctly. Optionally check chktex and pdfcrop are working.

4. Set up your OpenAI API key:

   export OPENAI_API_KEY='your-api-key'

Once installed, generate slides from any arXiv paper:

python paper2slides.py all 2505.18102

This will download the paper, generate slides, compile to PDF, and open the presentation automatically.

The paper2slides.py script provides a CLI interface with subcommands:

# Full pipeline (most common usage)
python paper2slides.py all <arxiv_id>

# Generate slides (beamer) only
python paper2slides.py generate <arxiv_id>

# Generate slides (beamer) with linting and PDF cropping
python paper2slides.py generate <arxiv_id> --use_linter --use_pdfcrop

# Compile slides (beamer) to PDF
python paper2slides.py compile <arxiv_id>

# Full pipeline without opening PDF
python paper2slides.py all <arxiv_id> --no-open

Replace <arxiv_id> with the desired arXiv paper ID. The ID can be identified from the URL: the ID for https://arxiv.org/abs/xxxx.xxxx is xxxx.xxxx.

The underlying tex2beamer.py and beamer2pdf.py scripts handle the core functionality:

• tex2beamer.py downloads and processes the arXiv paper using arxiv-to-prompt, then generates Beamer slides via OpenAI API
• beamer2pdf.py compiles the LaTeX slides to PDF using pdflatex

The prompts sent to the LLM and responses are logged to tex2beamer.log. Linter output (when --use_linter is used) is saved to source/<arxiv_id>/linter.log.

The prompts are now managed through a YAML-based system in prompts/config.yaml. This file contains:

• Template variables: Common settings like num_slides, max_items, figure_width, etc.
• Stage management: Separate prompts for initial, update, and revise stages
• Default values: Configurable defaults for audience, formatting, and dimensions
• Variable substitution: Dynamic replacement of {variable} placeholders

You can customize the prompts by editing prompts/config.yaml. The system automatically handles figure path insertion and other dynamic content. The PromptManager class in prompts/manager.py handles template rendering and validation.

Run the Streamlit app for a point-and-click experience:

streamlit run app.py

• Enter an arXiv ID (e.g., 2505.18102) or a search query in the sidebar.
• Click "Run Full Pipeline" to generate slides and compile the PDF, or use "Generate Only" / "Compile Only".
• After the pipeline completes, you can interactively edit the slides in the left panel; changes are recompiled automatically.

The process begins by downloading the source files of an arXiv paper. The main LaTeX file is identified and flattened, merging all input files into a single document (FLATTENED.tex) with arxiv-to-prompt. We preprocess this merged file by removing comments and the appendix. This preprocessed file, along with instructions for creating good slides, forms the basis of our prompt.

One key idea is to use Beamer for slide creation, allowing us to stay entirely within the LaTeX ecosystem. This approach essentially turns the task into a summarization exercise: converting a long LaTeX paper into concise Beamer LaTeX. The LLM can infer the content of figures from their captions and include them in the slides, eliminating the need for vision capabilities.

To aid the LLM, we create a file called ADDITIONAL.tex, which contains all necessary packages, \newcommand definitions, and other LaTeX settings used in the paper. Including this file with \input{ADDITIONAL.tex} in the prompt shortens it and makes generating slides more reliable, particularly for theoretical papers with many custom commands.

The LLM generates Beamer code from the LaTeX source, but since the first run may have issues, we ask the LLM to self-inspect and refine the output. Optionally, a third step involves using a linter to check the generated code, with the results fed back to the LLM for further corrections (this linter step was inspired by The AI Scientist). Finally, the Beamer code is compiled into a PDF presentation using pdflatex.

The unified paper2slides.py script automates the entire process, typically completing in less than a few minutes with GPT-4.1 for a single paper.