PPT Master: The Open-Source Tool That Finally Makes AI Slide Decks Actually Editable
You’ve done this before. You found an AI tool online that claimed to make PowerPoint slides. You typed your content in, clicked generate, and watched it spit out a beautiful deck. You opened it in PowerPoint, clicked on a text box to fix a typo — and nothing happened. The “text box” was a flat image. A screenshot of a slide, not a real slide.
That moment of betrayal is exactly why PPT Master exists.
Built by Hugo He, a finance professional who spends his working life reviewing and editing hundreds of slide decks, PPT Master hit 9,300 GitHub stars in a matter of weeks. It does one thing that almost no other AI slide tool does: it outputs a PowerPoint file where every shape, every text box, every chart is a real PowerPoint object. Click anything. Edit anything. No conversion needed. No “Convert to Shape” workaround. Just a normal, fully editable .pptx file.
What Is PPT Master, Really?
Think of PPT Master like a very smart intern who reads your documents and turns them into slides — but instead of handing you a printout you can’t edit, they hand you the actual working file with all the layers intact.
You give it a source — a PDF report, a Word document, a webpage URL, or even raw text pasted into a chat — and it hands back a PowerPoint file you can open, click around in, and edit just like one you built yourself from scratch.
The key word is natively editable. Every element PPT Master creates is stored as real PowerPoint XML under the hood (the technical name is DrawingML, which is just the file format PowerPoint uses internally). This matters because every other AI slide tool typically generates a pretty picture of a slide and embeds that picture into your .pptx. You get something that looks like a slide but behaves like a photograph of one.
flowchart LR
A[You have a document\nPDF, URL, DOCX] --> B[PPT Master reads it\nvia AI chat]
B --> C[AI designs slides\nas SVG drawings]
C --> D[Scripts convert SVG\nto real PowerPoint shapes]
D --> E[You get a .pptx\nyou can click and edit]
Every step from document to editable slide happens automatically — you only talk to an AI chat window.
Why Does This Problem Even Exist?
Most AI slide tools work by generating a visual — basically a picture of what a slide should look like — and then embedding that picture into a PowerPoint container. It’s faster and simpler to build. The catch is that the user ends up with something that only looks like a PowerPoint but doesn’t behave like one.
Before PPT Master: You’d get a generated deck, try to fix a font size on slide 3, and realize you’d need to open a separate design tool, re-export, and re-import. Or you’d give up and keep the wrong font.
After PPT Master: You open the file. You double-click the text. You change it. Done. Exactly like every other file you’ve ever made in PowerPoint.
%%{init: {'flowchart': {'htmlLabels': true}} }%%
graph LR
subgraph Old_Way_Image_Export
X["AI generates<br/>a picture of a slide"] --> Y["Embedded as flat<br/>image in .pptx"]
Y --> Z["Cannot edit anything<br/>without redesigning"]
end
subgraph PPT_Master_Real_Shapes
P["AI designs layout<br/>as vector SVG"] --> Q["Converted to real<br/>DrawingML shapes"]
Q --> R["Click, edit, resize<br/>anything in PowerPoint"]
endThe difference between other tools and PPT Master is whether you can actually use what you get.
What’s Inside PPT Master — The Structure
PPT Master isn’t just a script you run. It’s a full workflow system with several moving parts, each doing a specific job. Understanding the structure helps you know where to look when things go well (or sideways).
graph TD
ROOT[PPT Master Repository] --> A[skills/ppt-master/]
ROOT --> B[scripts/]
ROOT --> C[templates/]
ROOT --> D[examples/]
ROOT --> E[docs/]
A --> A1[SKILL.md — the master workflow\nthe AI reads this to know what to do]
B --> B1[source_to_md/ — converts your PDF\nor URL to readable Markdown]
B --> B2[svg_to_pptx.py — turns AI drawings\ninto real PowerPoint shapes]
B --> B3[notes_to_audio.py — generates\nnarration from slide notes]
C --> C1[20 layout templates\n52 chart and diagram templates\n6700+ vector icons]
D --> D1[15 example projects\n229 real pages you can study]
The repository is organized so each folder has one job — design, conversion, templates, or examples.
The most important file in the whole repo is skills/ppt-master/SKILL.md. This is the instruction manual that the AI (Claude, Cursor, or whichever tool you use) reads before doing anything. If the AI ever loses track of what it’s supposed to do mid-session, you just tell it “read skills/ppt-master/SKILL.md” and it gets back on track.
The scripts/ folder is where the actual technical heavy lifting happens. Python scripts convert your source documents to a format the AI can read, and then convert the AI’s drawings back into real PowerPoint shapes.
How PPT Master Actually Works — Step by Step
PPT Master runs inside what are called AI IDEs (Integrated Development Environments) — tools like Claude Code, Cursor, or VS Code with Copilot. These are applications where you type to an AI and the AI can also run code on your computer. You don’t need to know how to code. The IDE is just the “place” where you have the conversation.
Here’s the full pipeline in plain language:
sequenceDiagram
participant You
participant AI Chat
participant Python Scripts
participant PowerPoint File
You->>AI Chat: Drop in a PDF or URL
AI Chat->>Python Scripts: Convert source to readable text
AI Chat->>AI Chat: Strategist plans slide structure and design
AI Chat->>AI Chat: Executor draws each slide as SVG
AI Chat->>Python Scripts: Run svg_to_pptx.py
Python Scripts->>PowerPoint File: Convert SVG shapes to DrawingML
PowerPoint File->>You: A .pptx with real, editable shapes
The AI plans and draws; Python scripts handle the technical conversion to PowerPoint format.
Step 1: Install Python. That’s the only technical requirement. Go to python.org, download Python, and during the install, check the box that says “Add to PATH.” Then open a terminal and run:
pip install -r requirements.txtThis installs everything else PPT Master needs automatically.
Step 2: Install an AI IDE. Claude Code (free to try), Cursor, or VS Code with GitHub Copilot all work. Claude Sonnet or Opus gives the best results — the AI needs to do precise visual layout math, and stronger models are noticeably better at it.
Step 3: Point the AI at your source material. Drop a PDF into the projects/ folder and tell the AI chat: “Create a PPT from projects/myreport.pdf.” Or paste text directly. Or give it a URL.
Step 4: Confirm the design spec. The AI will ask you one round of questions — how many slides, which template or free design, landscape or portrait. Answer these and it handles the rest.
Step 5: Wait and receive. Generation takes 10 to 20 minutes for a 10–15 page deck. It’s intentionally serial (one slide at a time) to keep the visual style consistent across slides. The output lands in exports/ — two files: a native .pptx you edit, and an _svg.pptx visual backup.
What You’ll Actually Get
These are the real outcomes people walk away with, not feature bullet points:
You’ll understand what “natively editable” means and why it matters. After using PPT Master once, you’ll immediately recognize which other AI tools are secretly giving you flat images. You can’t un-see it.
You’ll have a fully editable deck from a source you’d normally spend hours adapting. Turning a 30-page PDF research report into a 12-slide pitch deck by hand takes most people two to four hours. PPT Master does it in about 20 minutes.
You’ll know how to customize the output. The system supports 20 layout templates, 52 chart and diagram templates, and over 6,700 vector icons. You can also hand it any existing .pptx deck you like and say “replicate this as a template” — and it will extract the colors, fonts, and layouts to use going forward.
You’ll understand the cost structure. The tool itself is free. Your only cost is the AI model API. A complete presentation costs roughly $0.08 to $0.24 USD depending on which AI tool you use. VS Code Copilot at $10/month gives you approximately 100 full presentations.
Your data stays on your machine. The only thing that leaves your computer is the text sent to the AI model for processing — the files themselves never go to a third-party server.
Common Beginner Mistakes
You might pick a weak AI model to save money. The SVG layout generation requires the AI to calculate precise pixel coordinates, font metrics, and container sizes. Weaker models get this wrong — text overflows boxes, shapes misalign, charts look off. Claude Sonnet or Opus are the recommended starting points. Other models work, but the layout quality drops noticeably.
Here’s why that happens: PPT Master uses SVG (Scalable Vector Graphics — a drawing format where every element has exact X/Y coordinates) as an intermediate format before converting to PowerPoint. Drawing things at exact coordinates is a math-heavy task. Stronger models do this math more reliably.
You might skip reading the FAQ when something looks wrong. The repository has a detailed FAQ at docs/faq.md that covers the most common issues: why a specific page looks off, how to regenerate a single slide without redoing the whole deck, what to do when the AI loses context mid-session. Most problems are solved there in under two minutes.
You might expect perfection on the first export. The README itself says the generated .pptx is a high-quality starting point, not a final deliverable. Minor adjustments in PowerPoint — nudging a text box, tweaking a font — are expected and normal. The tool saves you 90% of the work; you finish the last 10% in PowerPoint.
Where to Go Next
Easy: Download the ZIP from GitHub (no Git required — just click Code → Download ZIP), follow the Windows or Mac install guide in the README, and run it with a document you already have. The first deck you generate tells you immediately whether this fits your workflow.
Medium: Explore the examples/ folder — 15 complete projects, 229 pages of real output. Study the templates in skills/ppt-master/templates/ and pick one that matches your usual presentation style. Set up the image generation backend (Gemini works best) to add AI-generated images to your decks.
Stretch: Try the voice narration feature. Add speaker notes to your slides, run notes_to_audio.py, and let the system generate narration in a cloned voice via ElevenLabs or MiniMax. Then let PowerPoint export the whole deck as an MP4 video — synced narration, transitions, and all. No video editing software required.
The thing that gets people is that this isn’t some clever hack or workaround. It’s a full rethinking of how AI should generate presentations — starting from the constraint that a slide file should actually behave like a slide file. Hugo He built it because he was tired of AI tools that ignored that constraint. Nine thousand stars later, it turns out a lot of other people were tired of it too.
Install Python. Run one command. Drop in a document. See what comes back.
Drop any document. Get back a PowerPoint you can actually click and edit.