Paper2Slides: Convert Papers to Presentations in One Click
Turn research papers, reports, and documents into professional slides and posters in minutes, with universal format support and custom styling.
Evidence shows: CLI requires user-provided API keys (.env), but least privilege is not explicitly addressed; no user confirmation mechanism; data flow is described in README (RAG indexing, checkpoints), but not how data is transmitted to external APIs; sensitive data handling (API keys) is hinted via .env.example, but no encryption or protection measures; dependencies are not fully pinned, no security audit mentioned; external effects include generating files and calling external APIs, but side effects not documented; rollback is supported via checkpoints but not explicitly; source attribution is clear via MIT license and HKUDS copyright, but publisher not verified. Deductions: lack of user confirmation, insufficient least privilege documentation, dependency security not addressed.
Evidence shows: README is consistent, CLI options match documentation; dependency list exists but not all versions pinned, affecting reproducibility; failure messages not documented. Deductions: dependency versions not fully pinned, failure handling documentation missing.
Evidence shows: clear target audience (researchers, report makers), diverse scenarios (papers, reports, multiple formats); capability boundaries via CLI options and modes (fast/normal); trigger precision via CLI parameters and --from-stage; environment fit has conda and pip instructions, but OS compatibility not mentioned. Deductions: incomplete environment fit info.
Evidence shows: README structure clear with TOC; install steps detailed; naming stable (CLI options consistent); examples and FAQ via COMMUNICATION.md link; known limitations not explicitly listed; MIT license clear; versioning changelog has two news entries but no formal changelog; maintenance responsibility not explicit. Deductions: missing known limitations, formal changelog, and maintenance responsibility.
Evidence shows: outputs are PDF and images, directly usable; high marginal value (automates slide generation); cost-benefit not discussed, but external API usage may incur costs. Deductions: cost-benefit analysis missing.
Evidence shows: feature claims in README lack test or verification evidence; no cross-source corroboration; facts and inferences not clearly separated. Deductions: lack of verifiable evidence.
- API key management: Users need to store keys in .env, but no secure storage guidance is provided, posing leakage risk.
- External API calls: Generation relies on external LLM and image generation APIs, which may incur costs and data transmission; users should be aware of privacy.
- Dependency security: Dependencies are not fully pinned and no security audit is mentioned, posing supply chain risks.
What does this agent do, and when should you use it?
Paper2Slides is an open-source tool from HKUDS that automates presentation generation from documents using a four-stage pipeline (RAG indexing, content analysis, layout planning, and visual generation). It accepts PDF, Word, Excel, PowerPoint, Markdown, and more, offering both CLI and web interfaces. Users can choose built-in themes or describe custom styles in natural language, and checkpointing enables resumable progress. Requires Python 3.12 and API keys for LLM and image generation services.
Paper2Slides reads document files (PDF, Word, Excel, PowerPoint, Markdown, etc.), parses them using RAG (leveraging LightRAG and RAG-Anything) to build a searchable index, extracts figures, tables, and structure, then plans slide/poster layouts and invokes image generation models (supports OpenRouter or Google Gemini) to create visuals, finally outputting PNG and PDF files. The CLI offers options like --input, --output, --style, --length, --fast, --parallel, and saves checkpoints for resumability. The web interface uses FastAPI backend and React frontend, launched via start.sh.
- Researchers converting a lengthy paper into conference slides without manual formatting.
- Educators transforming lecture notes into engaging presentation decks with custom themes.
- Product managers turning market reports into internal pitch decks, iterating on styles quickly.
- Content creators generating posters from PDFs or Word docs for conferences or social media.
- Teams processing multiple documents in parallel to accelerate slide production.
- Users recovering from interruptions using checkpoints to resume without restarting.
What are this agent's strengths and limitations?
- Supports multiple document formats (PDF, Word, Excel, PPT, Markdown) and simultaneous processing.
- Four-stage pipeline with RAG ensures accurate content extraction and source traceability.
- Checkpoint mechanism allows resuming from any stage and rapid style iteration.
- Custom styling via natural language descriptions offers high flexibility.
- Provides both CLI and web interfaces, catering to different user preferences.
- Relies on external LLM and image generation APIs, requiring API keys and incurring costs.
- Image generation defaults to a specific model (e.g., gemini-3-pro-image-preview) which may add latency and cost.
- Requires Python 3.12 environment and possibly specific system dependencies.
- RAG indexing can be slow for long documents; fast mode compromises content extraction depth.
- Project is very new (open-sourced Dec 2025) with limited community maturity and potential maintenance risks.
How do you install or deploy this agent?
- Clone the repo:
git clone https://github.com/HKUDS/Paper2Slides.gitandcd Paper2Slides. 2. Create conda environment:conda create -n paper2slides python=3.12 -yand activate. 3. Install dependencies:pip install -r requirements.txt. 4. Create a.envfile in thepaper2slides/directory with your API keys (see.env.example).
How do you use this agent?
CLI example: python -m paper2slides --input paper.pdf --output slides --style doraemon --length medium --fast --parallel 2. Options include --input, --output (slides or poster), --style (academic/doraemon/custom text), --fast (skip RAG indexing), --parallel (parallel generation). For web UI, run ./scripts/start.sh and visit http://localhost:5173. To resume interrupted runs, just run the same command again; checkpoints auto-detect.
FAQ
What API keys are required?
.env.example in the paper2slides directory.How to resume an interrupted generation?
--from-stage to force restart from a specific stage.Can I customize slide styles?
--style with built-in themes (e.g., academic, doraemon) or any natural language description (e.g., 'Studio Ghibli anime style').What input formats are supported?
Difference between fast and normal modes?
--fast) skips RAG indexing, using direct LLM queries for short documents and quick previews. Normal mode builds a full RAG index for long or complex documents.