PPT Master is an agent skill that turns source documents into PowerPoint decks built from native PowerPoint objects. Rather than flat images or text boxes dropped into a template, the output uses native shapes and connectors with adjustment handles, data-backed charts and tables, the full text, picture, fill, and effect model, and, through the template route, real slide masters and layouts. Every element stays editable in PowerPoint.
It runs inside any agent-capable AI tool that can read and write files, run commands, and hold a multi-turn conversation: IDE agents such as Cursor and Windsurf, extensions such as Claude Code and GitHub Copilot, or CLI agents such as Codex CLI and Gemini CLI. You point the agent at a PDF, DOCX, web page, or pasted text and ask for a deck. By default the agent first confirms a design spec (template, format, page count), then analyses the content, structures the argument, designs each page as SVG, and exports a PPTX. A quick mode skips the confirmation and generates in one pass.
Apart from calls to the AI model, the pipeline runs on your machine. Beyond generating new decks, it distills reusable brand, style, and layout templates from reference decks, fills an existing PPTX with new content while keeping its design, and adds native transitions, animations, speaker notes, and audio narration to a finished deck. SmartArt is deliberately out of scope.
The model sets the quality ceiling: the project recommends Kimi K3 or Claude with a large context window, plus an image-generation model for illustrations. Claude, GPT, Gemini, Kimi, and other models can drive the workflow.
Features
- Native PowerPoint output: shapes, connectors, charts, tables, and formulas as editable DrawingML and OMML objects, not images
- Data-backed charts: an option exports charts and tables as native PowerPoint Chart and Table objects with Edit Data support
- Template routes: distills brand, style, layout, and deck templates from references and builds decks on real slide masters and layouts
- Edit existing decks: fills a supplied PPTX with new content, keeps its design and unchanged pages, and supports page selection and reordering
- Animation and narration: adds native transitions, animations, speaker notes, and audio narration, including voice cloning, to a finished deck
- Source formats: PDF, DOCX, HTML, EPUB, notebooks, images, and pasted text, with Pandoc for legacy formats such as DOC, ODT, RTF, and LaTeX
- Image acquisition: AI image generation through the host agent or a configured provider, or web image search across Openverse, Wikimedia Commons, Pexels, and Pixabay with automatic attribution
- Design confirmation: confirms template, format, and page count before generating, or skips the step in quick mode
- Agent hosts: runs in IDE agents, editor extensions, and CLI agents, and installs as a Claude Code plugin or a cross-agent skill
