Presentation decks built as image mocks, not template fill-ins

What it does

gpt-image2-ppt-skills is an agent skill for Claude Code, Codex, and similar tools that generates complete PowerPoint decks using OpenAI’s gpt-image-2. Instead of dropping text into rigid layouts, it treats every slide as a full visual mockup—generating 16:9 images that get packaged into a .pptx file. You can pick from 32 curated visual styles or upload an existing .pptx template and have the AI mimic its layout, color palette, and illustration style with entirely new content.

The interesting bit

The editing model is unusual: because each slide is rendered as a whole-page image, making changes means regenerating the target page via image-to-image prompts while the agent tries to hold the style steady. You sacrifice native PowerPoint editability for the ability to revise layouts by simply describing what you want changed.

Key highlights

• 32 built-in visual styles ranging from Swiss Grid and Editorial Mono to Y2K Chrome and Dark Aurora
• Template cloning: feed it any .pptx and it replicates the layout and aesthetic with your new content
• Natural-language editing: describe changes to specific pages or elements and the skill regenerates only those targets
• Dual output: every slide comes as a high-res PNG plus a packaged 16:9 .pptx ready to present
• Real-asset preservation: user-provided screenshots, logos, and charts embed as native objects by default instead of being redrawn by the image model

Caveats

• Text and backgrounds are whole-page images, so you cannot manually edit individual words or objects in PowerPoint afterward
• Dense tables, financial figures, and legal evidence require strict manual verification; the README explicitly warns against trusting generated numbers blindly
• Template cloning requires a local rendering backend such as PowerPoint, Keynote, or LibreOffice to analyze the source file

Verdict Worth a look if you generate decks via AI agents and care more about visual polish than native editability. Skip it if your workflow depends on tweaking individual text boxes, charts, or data tables by hand after generation.