PPT Master — AI-Generated Native PowerPoint Skill
Feed AI a document or topic and get a real, native PowerPoint: shapes, transitions, animations, data-backed charts, audio narration — running locally, your data stays local.
Extremely long doc (100k+ chars); only the opening was read. The process is written with strict phase gates, but generation quality wasn't verified by actually running it.
Static read only, no execution to verify output quality.
- The document is very long; this review did not read it end to end, so deep edge-case handling is unverified
What does this skill do, and when should you use it?
PPT Master is a free, open-source (MIT) workflow skill by Hugo He that runs inside any agent-capable AI tool such as Claude Code, Codex, or Gemini CLI. It turns PDFs, DOCX files, web pages, or pasted content into natively editable .pptx decks with slide masters, native shapes and connectors, on-demand data-backed charts and tables, native transitions with opt-in animations, and audio narration generated from speaker notes. Beyond generating new decks, it fills your existing .pptx templates with new content, distills reusable Brand/Style/Layout templates from references, and beautifies or adds narration to finished decks. The pipeline runs entirely on your machine; the only cost is your AI model usage.
Reads user-supplied PDF, DOCX, images, or pasted text, confirms a design spec (template, aspect ratio, page count), then generates pages and exports an editable .pptx. Technically it generates SVG first, then a converter writes native DrawingML objects; by default charts/tables export as editable SVG-derived shapes, and --native-charts-and-tables swaps eligible groups for PowerPoint-native Chart/Table objects backed by data. Other routes: fill an existing .pptx template preserving its design, create Brand/Style/Layout/Master template workspaces, and edit, beautify, or add transitions, animations, and notes-based narration to finished files. Image acquisition supports AI generation (image_gen.py with OPENAI_API_KEY/GEMINI_API_KEY etc.) and zero-config web image search (image_search.py, optional Pexels/Pixabay keys, automatic licensing and attribution). Each run passes scripts/attribution_guard.py integrity check, then routes through workflows/routing.md to exactly one workflow.
- Professionals who regularly review and edit presentations and want AI slides to stay editable in PowerPoint rather than flattened into images (the author himself is a finance CPA).
- Employees or students needing to quickly turn long reports, papers, or web pages into a deck they can keep polishing.
- Users with an existing corporate .pptx template who want AI to fill new content while preserving the design.
- Lecturers and trainers wanting narration audio or self-running video added to a finished deck for courses or presentations.
- Design or brand leads distilling reusable brand/layout templates from strong reference decks.
- Analysts producing chart-driven business or data-journalism decks with real, editable chart data.
What are this skill's strengths and limitations?
- Outputs real native PowerPoint objects — masters, shapes with adjustment handles, data-backed charts/tables — not flat images, so you keep editing in PowerPoint
- Pipeline runs locally; apart from model communication your data stays on your machine, with no platform lock-in
- Free and open source (MIT), no subscription — the only cost is AI model usage
- Multiple routed workflows: template fill, template distillation, beautify, edit, narration
- Not tied to one AI vendor — any agent-capable tool with file/command access can drive it
- Strict process discipline: blocking user-confirmation gates and an attribution_guard.py integrity check
- Quality is model-dependent: the author explicitly recommends a large-context (~1M token) Kimi K3 or Claude plus gpt-image-2 or gemini-3.1-flash-image for images; cheaper models yield a noticeable quality gap and more manual polishing
- Requires local Python setup and running scripts — a real hurdle for non-technical users (Windows needs extra steps)
- SmartArt is a deliberate omission, an acknowledged gap versus PowerPoint
- --native-charts-and-tables output may render differently across applications
- No test suite or independent quality evaluation is provided in the source material; real-world results vary by setup
How do you install this skill?
Prerequisite: Python 3.10+ (on Windows, check "Add to PATH" during install). Install: git clone https://github.com/hugohe3/ppt-master.git && cd ppt-master && pip install -r requirements.txt. Alternatives: download the ZIP (quick trial), or install via skill marketplace: npx skills add hugohe3/ppt-master, or in Claude Code: /plugin marketplace add hugohe3/ppt-master && /plugin install ppt-master@ppt-master (after marketplace install you still must run pip install -r requirements.txt from the installed location). Pandoc is only needed for legacy formats (.doc/.odt/.rtf etc.). Keys: for clones, cp .env.example .env; for skill installs, use ~/.ppt-master/.env.
How do you use this skill?
Open the ppt-master folder in an agent-capable tool (Claude Code, Cursor, Codex CLI, Gemini CLI, etc.), place source files in projects/ or paste content in chat, e.g.: "Please create a PPT from projects/q3-report/sources/report.pdf". The AI confirms the design spec, then generates and exports the .pptx. To reuse a template: give the AI your .pptx plus new material and say "fill this deck with the new content". For narration: ask the AI to add notes-based narration to a finished deck. If the AI loses context, ask it to read skills/ppt-master/SKILL.md; for other issues see docs/faq.md.
How does this skill compare with similar options?
The source names no specific competitors, but positions itself against typical "editable slides" tools by emphasizing native PowerPoint object-model depth (masters, adjustable shapes, real data charts) rather than editable text boxes alone, and by committing to keep converging with PowerPoint's own capabilities.