hugohe3/ppt-master

PPT Master is a workflow, packaged as an agent "skill," that lets an AI chat tool read your source PDFs, DOCX files, text or images and export a natively editable .pptx. It suits people who want PowerPoint objects they can keep editing, rather than flattened images or a filled-in template, and who are comfortable installing Python and driving an agent-capable AI tool.

What changes in the workflow

You put source material in a project folder and ask the agent in chat to build a deck; the agent handles content analysis, visual design, SVG generation and PPTX export. The default flow first confirms a design spec (template choice, format, page count), while an explicitly requested quick mode skips that round trip. The quick path is documented as one-pass and non-resumable, with no svg_final/ preview. Separately from generating new decks, you can also distill reusable templates from reference material, fill an existing .pptx while preserving its design, and add transitions, animations or narration to a finished deck — each route documented with its own preservation contract Getting Started.

Why the output is different

Rather than placing images or filling a template, it writes DrawingML with native shapes, connectors, charts and tables, and can export data-backed PowerPoint Chart/Table objects with Edit Data into a separate _native_charts_tables.pptx; formulas compile to editable OMML for PowerPoint 2010+. Through the template route it can produce real slide masters and layouts. The author frames this as an ongoing direction rather than a complete feature checklist and states that SmartArt is a deliberate omission Product Positioning. Example decks, described as single-pass with no manual polish, are browsable online Live Demo.

Requirements and limits

• Setup: Python 3.10–3.14 plus pip install -r requirements.txt, and an AI tool with file read/write, command execution and multi-turn chat. Claude Code is the author's recommended environment; the README also lists IDE agents, plugins and CLI agents as options. Git is recommended for updates but a ZIP download works; Windows needs extra PATH/execution-policy steps README.
• Model quality sets the ceiling: the author recommends a large-context model (around 1M tokens) plus AI image generation and warns of a quality gap with other models. The result is not expected to be a finished deck in one pass; polishing remains with you.
• Cost and data: the project is MIT-licensed and free, with only AI model usage to pay for. Apart from chosen model, search, image and speech services, the pipeline runs locally.
• Optional extras: image search works without keys, but Pexels/Pixabay keys improve stock-photo quality, and attribution credits are added when a license requires it. Pandoc is only needed for legacy formats. The optional PDF converter depends on PyMuPDF under AGPL-3.0, which matters if you redistribute a bundle containing it.

HotRepo's take

The repository is MIT-licensed and has drawn rapid star growth in the tracked board sample, but those signals show attention, not deck quality or reliability. The concrete decision is whether an agent-driven, Python-based workflow that produces editable native PowerPoint fits your setup — and whether the model access you have is strong enough for the quality you need.