agy-pptx-studio

Read this in other languages: 繁體中文

"agy-pptx-studio is not just a formatting tool; it's a highly modular, aesthetic-driven automated presentation production line."

Important

🚀 Public Beta Notice This ecosystem is currently in the Public Beta phase. Some multi-agent workflows and image generation boundaries are subject to change in future releases.

"About agy-pptx-studio"

Welcome to agy-pptx-studio, a dedicated ecosystem and monorepo designed specifically for AI-driven, highly aesthetic presentation (PPTX) generation. This repository centralizes tools, skills, and plugins necessary to construct, style, and generate rich slide decks using Antigravity AI agents. By utilizing Domain-Driven Design (DDD), this studio ensures all features within the ecosystem are cohesive and optimized for presentation creation.

1. 🤖 Multi-Agent Workflow (Presentation Architect)

This core plugin is structured as a Multi-Agent Department, acting as the brain for generating highly aesthetic presentations. The workflow enforces a strict sequential pipeline, coordinated by the Orchestrator, which receives the initial user command and delegates tasks:

workflow: [User Input] ➔ 🔍 Researcher ➔ 📝 Strategist ➔ 🎨 Art Director ➔ 🛠️ Engineer ➔ 👁️ QA Tester ➔ [Output PPTX]

• 🔍 Researcher (presentation-researcher): Begins by gathering raw data, performing deep analysis, and extracting necessary insights, saving them to a physical research file.
• 📝 Strategist (presentation-strategist): Constructs a logical, compelling narrative storyboard. It determines the flow of the slides and outputs a foundational outline.
• 🎨 Art Director (presentation-art-director): Applies design system constraints, selects typography and color palettes, invokes the image generator for necessary visuals, and finalizes the strict _blueprint.yaml.
• 🛠️ Engineer (presentation-engineer): Executes the blueprint using the pptxgenjs Node.js library to render the final, physical .pptx file.
• 👁️ QA Tester (presentation-qa-tester): Acts as the final gatekeeper, introducing a Visual TDD (Test-Driven Development) loop. It takes screenshots of the generated slides and uses a visual model to audit layout aesthetics. If text overflow or alignment issues are detected, it creates an Internal Ticket and dynamically routes it back to the responsible subagent for a closed-loop correction.

2. 🎨 Style Registry Architecture (Scheme D)

This project operates on a decentralized Style Registry Architecture to guarantee absolute layout stability and unlimited expandability. Instead of having the AI calculate x, y coordinates and guess layer orderings (which often results in broken designs), each visual theme is fully encapsulated into a self-contained module in the styles/ directory:

• Style Documentation (.md): A strict guideline for the AI Strategist and Art Director. It defines the exact layout_type IDs available for that specific style (e.g., S01_Cover, S02_Split_Text), ensuring the LLM only outputs supported layouts.
• Style Engine (.js): A precise pptxgenjs rendering script (e.g., styles/swiss-simple.js) that handles the exact coordinate mapping, typography, and collision prevention for that specific theme.

3. 📝 Speaker Notes & Typography Control

To prevent AI from cluttering slide layouts with unnecessary explanations or bilingual translations (e.g., Generative UI (生成式介面)), the ecosystem features a strict Bilingual Mandate:

• Pure Titles: Slide titles must be pure and punchy (either all English or all Chinese).
• Speaker Notes System: All supplementary context, explanations, and Chinese translations for technical terms are automatically routed to the slide's Speaker Notes (> Speaker Notes:). This keeps the visual layout extremely clean while preserving the AI's deep context for the presenter.
• Orchestration: A central scripts/build_presentation.js engine reads the AI's YAML blueprint, dynamically loads the correct Style Engine, injects the speaker notes, and renders the flawless .pptx.

4. 🖼️ Image Generation Enhancements

This module serves as the visual asset creator. It heavily modifies and enhances Antigravity's built-in generate_image command, overcoming native limitations to produce production-ready assets.

• Zero-Text / Zero-Subject Backgrounds: The native image generation often creates messy images. Our background-generation-formula enforces strict rules (NO TEXT, NO WORD) to generate clean, highly-readable backgrounds and UI overlays.
• Prompt Enhancer: Uses a 7-layer structural formula for general cinematic images, and a 5-layer formula (diagram-generation-formula) tailored for conceptual visualizations like timelines, funnels, and process cycles, covering 100% of 20 common presentation layouts.
• Native Aspect Ratio & Robust Fallback: Fully supports the native AspectRatio API parameter (e.g., 16:9, 9:16) for optimal composition quality. Simultaneously retains the clever "padding prompt" technique combined with an automated Python script (crop_image.py) as a fallback mechanism, ensuring the workflow never breaks if the native generator encounters limitations.

🛠️ Tech Stack

• Agent Framework: Antigravity, LangChain (Skills & Agents)
• Presentation Rendering: Node.js (pptxgenjs)
• Image Processing: Python (Pillow / Image Cropping scripts)
• Design Systems: YAML, JSON (Decoupled layout engines)

📂 Architecture Directory

agy-pptx-studio/
├── presentation_architect/      # Core multi-agent presentation generation plugin
│   ├── plugin.json
│   ├── skills/                  # Individual subagent skills (Strategist, Art Director, Engineer)
│   ├── scripts/                 # Node.js build orchestrator (build_presentation.js)
│   └── styles/                  # Style Registry containing .md guidelines and .js engines (e.g., swiss.js)
└── antigravity-image-master/    # Visual asset & image generation plugin
    ├── plugin.json
    └── skills/                  # Prompt formulas and ratio cropping tools

1. Installation

Clone this project and link the plugins to your Antigravity configuration:

git clone https://github.com/YuJunWang/agy-pptx-studio.git

(Place or symlink the two plugin folders into your Antigravity .gemini/config/plugins directory to enable them in your workspace.)

2. Usage Example

To trigger the complete multi-agent workflow, you can simply use a comprehensive prompt like this:

"Based on coffee industry research from the past two years, act as a local coffee supplier in Taiwan and create a 15-minute pitch deck for venture capitalists. Execute this in a McKinsey consulting presentation style, including as much data and as many charts as necessary. You can decide on any other unspecified details."

👨‍💻 Author

Yu-Jun Wang

• GitHub Profile

📄 License

This project is licensed under the MIT License.