JuneYaooo/gpt-image2-ppt-skills

What it solves

This project provides a way to generate high-aesthetic, professional PowerPoint presentations using OpenAI's gpt-image-2 model. It moves away from traditional "template filling" by treating each slide as a complete visual composition, ensuring a unified and polished look from cover to content pages.

How it works

The tool operates as a "Skill" for AI agents (such as Claude Code, Codex, and others). It uses natural language prompts to generate 16:9 high-definition images for each slide and packages them into a .pptx file. It supports two primary modes:

• Visual-First Mode (Default): Generates full-page visual compositions to maximize the model's aesthetic and layout capabilities.
• Editable Mode (Optional): When explicitly requested, the tool reconstructs the visual composition into native PowerPoint objects, including editable text, basic shapes, connectors, and independent image layers.

It also features a "Template Clone" mode where it analyzes a provided .pptx or image template to replicate its layout, color scheme, and visual language for new content.

Who it’s for

Users who need professional-looking presentations for reports, roadshows, training, and product introductions, especially those who want to avoid the tedious process of manual layout and design.

Highlights

• Template Cloning: Replicates the visual style of an uploaded .pptx file to create new content.
• Editable Output: Ability to convert visual slides into native PowerPoint objects for manual refinement.
• Editable Gallery: Access to a curated online gallery of templates with prompts that can be used directly with AI agents.
• Precise Natural Language Editing: Allows users to modify specific elements (e.g., "change the subtitle on page 3") without regenerating the entire deck.
• High Efficiency: Supports parallel generation of up to 10 images to produce a 10-page deck in approximately 30 seconds.
• Dual Output: Provides both high-definition PNGs and a packaged .pptx file.