What you’ll need

  • Python 3.10+ — the only hard install. Everything else comes from one pip install -r requirements.txt.
  • An agent-capable AI tool — any chat that can read and write files, run commands, and sustain a multi-turn conversation: Claude Code (the author’s pick, and the environment the project is tested on most), Codex CLI, Gemini CLI, Cursor, Cline, or an IDE extension with an agent panel.
  • A model key — PPT Master itself is free and open source (MIT); the only cost is your model usage. The recommended pairing is Kimi K3 (or Claude) with a large (~1M token) context window, plus an image model (gpt-image-2 or Google’s gemini-nano-banana-2.1) if you want generated imagery.
  • Pandoc — the document-conversion backend: sudo apt install pandoc on Debian/Ubuntu, brew install pandoc on macOS. Windows users get a dedicated step-by-step guide (check “Add to PATH” when installing Python).

1. Clone the project and install dependencies

The recommended path is a git clone, so you can pull updates later:

git clone https://github.com/hugohe3/ppt-master.git
cd ppt-master
pip install -r requirements.txt

No git? Click Code → Download ZIP on the GitHub page, unzip, and run the same pip install -r requirements.txt. If the full download is too heavy, grab the skill-only package ppt-master-skill-*.zip (about 56 MB, fully functional but without the bundled example decks) from the Releases page.

2. Install it as a skill in your AI tool

PPT Master is a workflow (a “skill”) that runs inside your agent. From any cross-agent CLI:

npx skills add hugohe3/ppt-master

Or inside Claude Code:

/plugin marketplace add hugohe3/ppt-master
/plugin install ppt-master@ppt-master

Then point the agent at the project folder: cd ppt-master before launching a CLI agent, or File → Open Folder in an IDE. Everything from here on happens in chat — you never write code.

3. Drop in your material and generate the deck

Put your PDF, DOCX, or images in the projects/ directory, then tell the AI chat which files to use. The quickest way to get the path: right-click the file in your file manager or IDE sidebar, copy the path, and paste it into the chat:

Please create a PPT from projects/q3-report/sources/report.pdf

You can also paste text straight into the chat and it will build a deck from that. By default the AI first confirms the design spec with you — template, 16:9 format, page count. To skip the round trip, say so explicitly and it goes straight to authoring:

Quickly generate a 5-page deck from projects/q3-report/sources/report.pdf — no need to confirm with me

The deck lands in exports/<name>_<timestamp>.pptx as natively editable DrawingML, and the default flow also writes self-contained page previews to svg_final/ so you can review without opening PowerPoint.

4. Level up: native charts, images, and your own templates

An AI agent turning a document into slides in a chat window
Generation happens in chat: the agent analyses the source, reasons the argument into shape, and exports the deck. AI-generated illustration for AI Frontier Post.

The core pipeline is one route; three more are worth knowing:

  • Data-backed charts and tables. Charts ship as editable shapes by default; add --native-charts-and-tables to get real PowerPoint Chart / Table objects with Edit Data, saved as a separate _native_charts_tables.pptx.
  • Images. The agent can use its host’s native image tool, or python3 skills/ppt-master/scripts/image_gen.py --list-backends shows configured provider backends. For stock, image_search.py works with zero config (Openverse / Wikimedia Commons), and free PEXELS_API_KEY / PIXABAY_API_KEY values in .env materially improve results — start from cp .env.example .env.
  • Your brand, your decks. PPT Master can distill reusable templates from your reference decks, fill an existing .pptx with new content while keeping its design byte-for-byte, and add native transitions, animations, and narration to a finished deck.

What you built

A local document-to-deck pipeline: source in, real .pptx out. The selling point is depth in PowerPoint’s native object model — native shapes with working adjustment handles, data-backed charts and tables, real slide masters and layouts, text that stays editable, formulas compiled to editable OMML (PowerPoint 2010+). Click any element and keep editing it like you built it by hand. Updating later is one script:

python3 skills/ppt-master/scripts/update_repo.py

Honest limitations

An editable slide with selection handles on a native chart
Everything stays native and editable — the whole point is a deck you can keep refining, not a flattened image. AI-generated illustration for AI Frontier Post.
  • The model sets the ceiling. PPT Master owns the workflow; the agent’s model owns the taste. Cheaper models mean more cleanup — if results disappoint, upgrade the model first, then check your usage against the Getting Started guide and the example projects.
  • Don’t expect a perfect deck in one shot. The value is removing the tedious 80%; the polishing left is yours — a natively editable deck exists precisely so you can keep working on it.
  • SmartArt is a deliberate omission, not a gap — the project’s PowerPoint ↔ SVG mapping guide is the honest record of what the native pipeline covers today.
  • Data stays local apart from model communication — but every generation bills your model provider, so budget it like any API usage.

Hugo He, a finance professional who reviews decks for a living, built PPT Master for one reason: AI-generated slides kept coming back as uneditable images. The project has since passed 60,000 GitHub stars. Now you can stop wrestling with flattened exports and start from a deck that’s yours to edit.