PPT Visual Reconstruction

The strongest practical workflow for turning slide images into editable PPTX decks.

PPT Visual Reconstruction is a Codex skill for high-fidelity image-to-editable-PowerPoint work. It is built for the hard cases: screenshots, one-picture-per-slide decks, AI-generated slide images, visual mockups, dense Chinese slides, textured backgrounds, icons, product UI, medical or technical visuals, and designs where naive OCR-to-PPT destroys the look.

Why This Exists

Most image-to-PPT tools choose one bad extreme:

• flatten the whole slide as one image, so nothing useful is editable;
• OCR everything into basic boxes, so the final deck looks obviously rebuilt.

This skill uses a hybrid reconstruction strategy:

• keep complex visuals as high-quality bitmap layers;
• remove old bitmap text only where needed;
• rebuild titles, labels, callouts, page numbers, footers, simple shapes, rules, and key copy as editable PowerPoint objects;
• render and compare the result before delivery.

That gives the result people actually need: a deck that still looks like the source, while the important text and structure can be edited.

Best For

• full-slide PNG/JPG screenshots;
• PPTX files where each slide is mostly one large image;
• AI-generated slide images that need to become editable;
• Chinese business, medical, technical, report, dashboard, and conference slides;
• slides with complex backgrounds where text removal must not leave obvious flat blocks.

Install

After this repo is published, install it into Codex with:

python "$env:USERPROFILE\.codex\skills\.system\skill-installer\scripts\install-skill-from-github.py" --repo lirunjie0510/ppt-visual-reconstruction --path . --name ppt-visual-reconstruction

Manual install:

Copy-Item -Recurse ".\ppt-visual-reconstruction" "$env:USERPROFILE\.codex\skills\ppt-visual-reconstruction"

Restart Codex if the skill does not appear immediately.

Use

Invoke it explicitly:

$ppt-visual-reconstruction

把这个图片型 PPT / PPT 截图重建成高保真可编辑 PPTX。复杂背景、图标、产品图和纹理保留为高清图片层；标题、正文、标签、页码、脚注、结论框和简单形状尽量做成可编辑 PowerPoint 对象。输出 final PPTX、预览图、对比图和验证结果。

Or let Codex trigger it automatically when you ask for:

• image to editable PPTX;
• screenshot to PPT;
• bitmap slide to editable PowerPoint;
• restore / reconstruct a visual PPT;
• make text editable while preserving the source look.

Demo

The goal is not to flatten the slide. The output keeps the visual layer faithful while making the main information layer editable inside PowerPoint.

Before: source slide image	After: editable PPTX preview

This example shows editable title, subtitle, icon labels, footer text, and simple layout elements over a preserved high-fidelity visual background.

Why Star This Repo

If this project reaches thousands or tens of thousands of stars, it becomes more than a personal prompt pack:

• Trust signal: people looking for image-to-editable-PPTX workflows will try this first.
• Distribution: stars help the repo rank in GitHub search, newsletters, tool lists, and AI office automation communities.
• Commercial leads: agencies, designers, consultants, and enterprise users may ask for custom reconstruction, paid workflows, private templates, or batch automation.
• Product path: the same workflow can grow into a desktop tool, SaaS API, Codex plugin, template marketplace, or paid service.
• Personal brand: a high-star repo is strong proof that you can build practical AI office automation systems.

Star growth does not directly pay money by itself. The value is attention, credibility, and inbound demand. The money comes from converting that demand into services, products, licenses, consulting, or training.

Roadmap

• Add reproducible demo cases with original images, rebuilt PPTX files, and comparison previews.
• Add helper scripts for extracting full-slide images from PPTX files.
• Add a QA script that checks slide count, media health, and editable text strings.
• Add before/after gallery images for GitHub social sharing.
• Package the skill as a Codex plugin after the workflow stabilizes.

Repository Layout

ppt-visual-reconstruction/
+-- SKILL.md
+-- agents/
|   +-- openai.yaml
+-- references/
    +-- layering-and-qa.md

Positioning

This project is positioned as the current strongest practical solution for image-to-editable-PPTX reconstruction inside Codex-style agent workflows: not just OCR, not just SVG conversion, and not just a flattened screenshot, but a deliverable-focused reconstruction process that balances editability and visual fidelity.

If this helps your slide reconstruction work, star the repo so more people can find it.