SlidesAI

Give this tool a PDF or text file, it automatically extracts the content and builds beautiful PDF slides or a single-page academic poster.

Demos: OpenAI Pitch Deck (source) · Attention Is All You Need (source)

designer	editorial	midnight
blush	tech	premium
terra	slate	crimson

premium

terra

• Smart Extraction: Pulls text, tables, LaTeX equations, and figures from PDFs automatically.
• LLM-Powered Writing: Works with Gemini, OpenAI, Claude, and OpenRouter. Adjustable verbosity: concise, normal, or detailed.
• Academic & General Purpose: Built for research and technical content, but works for any topic or industry.
• 9 Handcrafted Themes: Professionally designed styles with curated typography and color palettes (see below).
• Academic Poster Generator: Add --poster to produce a print-ready single-page HTML poster (A0 landscape) instead of a slide deck.

1. Clone and install everything in one command:

git clone https://github.com/yourusername/slidesai.git
cd slidesai
python3 install.py

This installs all Python dependencies and Marp CLI automatically. If npm is not found, the script will skip Marp and print instructions — you only need it for PDF rendering.

(Note: marker-pdf may require extra system dependencies for OCR on some platforms. Refer to their docs if you hit issues. Google Chrome or Chromium must be installed for PDF rendering.)

2. Setup your API Key:

cp project_secrets.py.example project_secrets.py

Set LLM_PROVIDER and the corresponding key in project_secrets.py. Only the key for your chosen provider is required:

(This file is gitignored to keep your keys safe.)

Just point build.py at your file and everything happens automatically:

# From a PDF (extracts → generates slides → renders PDF)
python3 build.py paper.pdf

# From a text or markdown file
python3 build.py notes.txt --theme midnight

# Control slide count and detail level
python3 build.py paper.pdf --theme designer --num_slides 15 --verbosity detailed

# Extract only specific pages, concise output
python3 build.py paper.pdf --theme slate --page_range 0-10 --num_slides 10 --verbosity concise

# Use a different LLM provider
python3 build.py paper.pdf --provider openai --model gpt-5.2
python3 build.py paper.pdf --provider anthropic

# Generate an academic poster (HTML) instead of slides
python3 build.py paper.pdf --poster
python3 build.py paper.pdf --poster --theme crimson

Your output files will appear in a folder named after your input file.

💡 Poster tip: Open the generated {input_name}_poster.html in any browser and use File → Print → Save as PDF (set paper size to A0, landscape) for a print-ready conference poster.

💡 Re-style without calling the LLM again: After your first build, a cached LLM response is saved. Re-run with --use_cached to try a different theme instantly — no API call needed:

# Slides
python3 build.py paper.pdf --theme midnight --use_cached
# Poster
python3 build.py paper.pdf --poster --theme crimson --use_cached

positional:
  input_file       Path to input file (.pdf, .txt, or .md)

options:
  -o, --output_dir   Output directory (default: folder named after input file)
  --poster           Generate a single-page HTML academic poster instead of slides
  --theme            Theme (default: premium)
  --num_slides       Target number of slides — slides only (default: 14-20)
  --verbosity        Slide text density: concise | normal | detailed — slides only (default: normal)
  --provider         LLM provider: google | openrouter | openai | anthropic (default: google)
  --model            Model name, e.g. gpt-5.2, claude-opus-4-6, gemini-3-pro-preview (default: provider's recommended model)
  --use_cached       Skip the LLM call and reuse the cached response (default: off)
  --page_range       Pages to extract from PDF, e.g. 0-5, 10 (default: all pages)
  --disable_ocr      Disable OCR for PDF extraction (default: OCR enabled)
  --skip_pdf         Skip the final Marp-to-PDF render step — slides only (default: off)

If you prefer finer control, you can run each step individually:

python3 extract_with_marker.py --pdf_path my_paper.pdf --output_dir my_project/

Generates a markdown file and an images/ folder in the output directory.

options:
  --pdf_path       Path to input PDF
  --output_dir     Directory for output assets (default: same directory as PDF)
  --page_range     Page range to extract, e.g. 0-5, 10, 12-14 (default: all pages)
  --disable_ocr    Disable OCR and rely on embedded PDF text (default: OCR enabled)

python3 build_slides.py \
    --input_file my_project/my_paper.md \
    --output_file my_project/my_presentation.md \
    --theme designer \
    --num_slides 15 \
    --verbosity normal

options:
  --input_file     Path to your extracted content markdown
  --output_file    Path to save the generated presentation markdown (default: {input_filename}_slides.md)
  --num_slides     Target number of slides (default: 14-20)
  --theme          Presentation theme (default: premium)
                   choices: designer | editorial | midnight | blush | tech | premium | terra | slate | crimson
  --verbosity      Slide text density: concise | normal | detailed (default: normal)
  --provider       LLM provider: google | openrouter | openai | anthropic (default: google)
  --model          Model name (default: provider's recommended model)
  --use_cached     Skip the LLM call and reuse a previously saved response (default: off)
  --no_page_numbers  Do not show page numbers on slides
  --poster         Generate an HTML academic poster instead of a slide deck

python3 build_slides.py \
    --input_file my_project/my_paper.md \
    --output_file my_project/my_poster.html \
    --theme premium \
    --poster

Opens directly in the browser. Use File → Print → Save as PDF (A0 landscape) for a print-ready conference poster. A llm_poster_raw.txt cache file is saved alongside for fast re-theming with --use_cached.

python3 render_marp_pdf.py my_project/my_presentation.md

Automatically finds your Marp installation and Chrome to output my_project/my_presentation.pdf.

options:
  input_md         Path to the Marp markdown file
  -o, --output     Output PDF path (default: input filename with .pdf extension)
  --browser        Browser engine for Marp: auto | chrome | edge | firefox (default: auto)

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

MIT License — free to use, modify, and distribute.