Tonight I discovered a project with an unexpectedly clever approach: using GPT to generate luxurious image-format PPTs, then “decompiling” them back into editable PPTX files. Previously, AI-generated PPTs were either templated or could only export images. This project takes a direct “generate image → parse → reassemble” path. The core idea is to decompose the PPT image into four layers: background, framework, icons, and text, then restore them by coordinates. It’s split into three independent skills: you can use just image generation, or just convert images to editable files. This solves the fundamental problem that AI-generated content is often not editable, and the results look quite luxurious.

GordenSun/GordenSuperPPTSkills

Source: https://github.com/GordenSun/GordenSuperPPTSkills

Gorden Super PPT Skills (Skill Pack)

The ultimate AI PPT game‑changer – the most powerful PPT Skill ever!

Use GPT to generate a luxurious image-format PPT, then convert it into a fully editable PPTX file.

The complete workflow has been split into 3 independent skills, which can be used or optimized separately:

• Only need an image-based PPT → GordenImagePPTGen
• Only need to convert images to editable → GordenImage2PPTX
• One-click “generate images first, then convert to editable” / unspecified → GordenSuperPPTSkill (A→B pipeline)

Showcase

1. GordenImagePPTGen – Generate an image-format PPT

   Example of an image-format PPT generated by GordenImagePPTGen

2. GordenImage2PPTX – Convert images to a fully editable PPTX file

   Example of an editable PPTX converted by GordenImage2PPTX

How to Use

For Codex only.

Step 1: Send the GitHub repository URL to Codex and let it install the skills.

Step 2: Use as needed. Recommended model: GPT 5.5, reasoning strength: Medium.

If you only want to generate an image-format PPT, use a prompt like:

Use the GordenImagePPTGen skill to generate an N‑page PPT. The content should be XXX. The PPT must be luxurious, information‑dense, and with complex layouts.

If you only want to convert an image PPT to an editable PPTX file, use a prompt like:

Convert the XXX.png file in the current folder into an editable PPT using GordenImage2PPTX. Follow the skill steps strictly.

Note: This skill is designed for Codex only, because it relies on GPT’s image generation and vision capabilities. In theory, Opus + GPT’s image generation API could also work, but no special adaptation has been made for that setup.

Converting images to editable PPTX files consumes significant quota – converting one image costs about 10% of a Plus subscription’s 5‑hour quota.

By default, the framework is rendered as a single overall image, but it also supports splitting into individual framework‑module images. Just tell Codex explicitly in the prompt.

How It Works

The core relies on GPT’s image generation and visual parsing capabilities. The general steps are: extract the background image, framework image, icons/decorations, and text from the PPT image one by one, then reassemble them by coordinates inside a PowerPoint file. To achieve a perfect result, many detailed checks and constraints have been added. During use, you can see the intermediate images GPT generates.

Background Image

Background image during the image‑to‑PPTX process

Framework Image

Framework image during the image‑to‑PPTX process

Icons & Decorations

Icons and decorations during the image‑to‑PPTX process

Installation (for AI)

Each skill directory is self‑contained (includes its own scripts/ and references/). Copy as needed:

# Codex (choose one or all)
cp -R GordenImagePPTGen   "${CODEX_HOME:-$HOME/.codex}/skills/GordenImagePPTGen"
cp -R GordenImage2PPTX    "${CODEX_HOME:-$HOME/.codex}/skills/GordenImage2PPTX"
cp -R GordenSuperPPTSkill "${CODEX_HOME:-$HOME/.codex}/skills/GordenSuperPPTSkill"

GordenSuperPPTSkill depends on the other two skills, so install them together (in the same skills directory / repository). The easiest way: copy the entire repository over.

Dependencies

pip3 install python-pptx pillow numpy

The image generation backend is auto‑resolved at runtime (Codex uses built‑in imagegen). See references/runtime-notes.md in each skill for details.

Directory Structure (this folder can be copied entirely to another Agent)

GordenSuperPPTSkills/
├── README.md                  ← This file (main entry)
├── GordenImagePPTGen/         ← Skill A: generate image PPT (with its own scripts/ references/ sample images/)
├── GordenImage2PPTX/          ← Skill B: image → editable pptx (with its own scripts/ references/)
└── GordenSuperPPTSkill/       ← Orchestration A→B (with its own references/; calls the two skills above)

Each skill directory is self‑contained. Copy the entire GordenPPTSuperSkills/ folder to the target Agent’s skills directory to use.

Credits & License

• Commercial use is allowed, but you must credit the GitHub source, or mention the author @Gorden Sun.
• If you want to join the PPT Skill discussion group, add me on WeChat: duge360
• Thanks to the LinuxDO (https://linux.do) community for support