Breaking

HKUDS Releases Open-Source Paper2Slides, Drawing 3,700 GitHub Stars

July 2, 2026 at 04:41 EDT

Open Source
AI Agents
Research & Papers

A project linked to the University of Hong Kong's Data Intelligence Lab has released Paper2Slides, an open-source tool that converts research papers into presentation slides and posters with one click. Using RAG-based content extraction, it generates decks in minutes from multiple document formats such as PDF and Word.

Dec 8, 2025 · HKUDS · University of Hong Kong

Paper2Slides: Research Papers Become Slides & Posters, Automatically

An open-source tool that turns PDFs — and Word, Excel, PowerPoint, or Markdown — into polished presentation slides and posters through a four-stage, RAG-powered pipeline.

stages in the generation pipeline

input formats: PDF, Word, Excel, PowerPoint, Markdown

parallel workers by default (--parallel, added Dec 9)

The Four-Stage Pipeline

STAGE 1

Build RAG Index

→

STAGE 2

Summarize & Extract Structure

→

STAGE 3

Plan Generation

→

STAGE 4

Image Generation

A checkpoint feature lets the workflow resume after an interruption.

Key Command Options

--output

slides / poster

--style

academic / doraemon / natural-language custom

--length

short / medium / long (slides)

--density

sparse / medium / dense (posters)

--fast

skip RAG for a fast preview

--parallel N

parallel generation (default 2)

SEEN AS AN ADVANTAGE

Key-point extraction + visualization in one step
RAG-backed accuracy for faithful summaries
Flexible style control, from academic presets to free-form prompts

POTENTIAL DRAWBACKS

Output quality varies with LLM dependence
Cost and rate limits of image-generation APIs
Complex equations and figures may be hard to reproduce faithfully

Runs locally or via Docker. A React frontend with a FastAPI backend, image generation through OpenRouter (Gemini-3-pro-image-preview) or the Google Gemini API, with a Python 3.12 conda environment recommended. As the release is still recent, real-world evaluation awaits further verification.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.