ppt-master

Project Overview

PPT Master turns documents, PDFs, web pages, images or a one-line topic into a natively editable .pptx file, with the whole pipeline running on your own machine: the AI analyses the content, designs the layout and writes SVG, and the Python scripts in the repository compile that SVG into DrawingML. It is not a standalone app but a skill plus script bundle under skills/ppt-master, driven by any AI tool that can read and write files, run commands and hold a multi-turn conversation. The result is a real deck with slide masters, native shapes, charts, tables and OMML formulas rather than flattened images, and every page stays editable in PowerPoint.

Item	Value
Form	Skill pack plus Python scripts inside an AI tool
Output	Natively editable .pptx, with svg_final/ page previews by default
Runtime	Python 3.10–3.14
License	MIT

Swap letting the AI draw pictures for letting the AI write real PowerPoint.

Supported Hosts

Host type	Examples	How to use
IDE-native agent	VS Code, Cursor, Trae, Windsurf, Zed	Open the project folder and type instructions in the side chat panel
IDE plugin	Claude Code extension, GitHub Copilot, Cline	Installed inside VS Code or a JetBrains host
CLI agent	Claude Code CLI, Codex CLI, Gemini CLI	Start it in the terminal after entering the project folder

Key Features

• Native PPTX export: output lands in exports/name_timestamp.pptx, where shapes, text boxes and connectors are real PowerPoint objects you can keep editing
• Document to outline: reads PDF, DOCX, HTML, EPUB and IPYNB directly, while doc, odt, rtf, tex, rst, org and typ need Pandoc installed
• Masters and layouts: decks inherit real p:sldMaster and p:sldLayout structures, and an existing company template can be applied
• Native charts and tables: with --native-charts-and-tables the deck gets Chart and Table objects with Edit Data, saved as a separate _native_charts_tables.pptx
• Formulas and motion: formulas compile to OMML for PowerPoint 2010 and later, and native transitions, animations and narration from speaker notes are supported
• Image sourcing: image_gen.py generates images through an AI backend, while image_search.py searches the web and adds attribution automatically for CC0 and CC BY style licenses
• Refill an existing deck: hand the AI a finished .pptx and it keeps the design, edits only the pages you pick and leaves untouched pages byte-identical
• Canvas presets: ships PPT 16:9, Xiaohongshu, WeChat and more than 10 canvas sizes

Quick Start

1. Install Python 3.10–3.14, and tick Add to PATH on Windows
2. Clone the repository, enter the folder and run pip install -r requirements.txt
3. Open the ppt-master folder in an AI tool, or start a CLI agent inside that folder
4. Put your PDF, DOCX or images under projects/ and name the path in chat, for example make an 8-page deck from projects/q3-report/sources/report.pdf
5. Confirm the template, aspect ratio and page count the AI proposes, or ask it to quick generate without confirmation
6. Collect the file from exports/, and find page previews in svg_final/ for the default flow
7. Add --native-charts-and-tables when you need data-backed charts, which saves a separate _native_charts_tables.pptx
8. Copy .env.example to .env and fill in PEXELS_API_KEY or PIXABAY_API_KEY for better stock images, while Openverse search works without keys

Target Audience

• People who present often at work
• Teams that reuse a corporate template
• Marketers producing decks in bulk
• Technical users comfortable with AI tools

Core Advantages

• Model cost only: the software is free and open source, and the only spend is model usage plus the search, image and speech services you choose to enable
• Local data: the pipeline runs on your machine, so source material is not uploaded to a third-party platform first
• No model lock-in: Claude, GPT, Gemini and Kimi all drive the same workflow, so switching tools does not mean rebuilding the project
• Close to native: capabilities are added against the feature-by-feature mapping in docs/powerpoint-svg-mapping.md instead of pasting page screenshots into slides

System Requirements

Item	Requirement
OS	Windows 10 and later, macOS, Linux
Python	3.10–3.14
Optional dependency	Pandoc, only for legacy formats such as doc, odt, rtf, tex, rst, org and typ
AI tool	An agent with file access, command execution and multi-turn chat
Model context	Around 1 million tokens recommended for long documents

Notes

• The model sets the ceiling: the skill only provides the workflow, so layout quality follows the model's context and taste and weaker models fall short on the same pipeline
• One pass, no resume: quick mode produces a deck in a single pass and cannot be resumed, and it writes no svg_final/ preview
• PDF component license differs: the optional PDF converter relies on PyMuPDF under AGPL-3.0, so redistributing a bundle that includes it must follow those terms
• Bring your own assets: fonts, brand templates and high-resolution images are not in the repository, so corporate templates must be placed under projects/ or the template folder
• Output keeps growing: every run adds a new timestamped file under exports/, so long-term use needs manual cleanup