Skip to content

About

A Claude Code / Codex skill that converts image-based PPTX, slide screenshots and scanned decks into truly editable PowerPoint: real text boxes, native shapes, OCR-measured layout.

Topics

Resources

Code of conduct

Contributing

Stars

53 stars

Watchers

2 watching

Forks

Repository files navigation

Slide Alchemy

English | 简体中文

GitHub stars License

Slide Alchemy is a skill for Codex and Claude Code that rebuilds image-based decks, slide screenshots, and scanned slides as editable PowerPoint files.

What gets rebuilt

Source element PowerPoint output
Body text, headings, and decorative lettering Native text boxes. Gradients, outlines, and shadows use native text effects. More complex effects are approximated while keeping the text editable.
Lines, borders, cards, and other simple geometry Individually editable PowerPoint shapes.
Icons and complex illustrations Regenerated PNGs that can be moved, resized, and replaced independently.
Real photographs, such as group and event photos Cropped directly from the source slide and inserted as separate pictures.
Slide backgrounds Clean backgrounds created with an image editing model that removes foreground content.

Quick start

After installation and configuration, attach the image-based PPTX or slide screenshots. Use $slide-alchemy in Codex or /slide-alchemy in Claude Code.

Default mode

Use slide-alchemy to convert this image-based deck into an editable PPTX.

Review the proposed background groups, then review the clean background previews.

Unattended mode

Use slide-alchemy to convert this image-based deck into an editable PPTX. Run unattended without intermediate confirmation.

Both modes deliver an editable PPTX and comparison previews.

Workflow

  1. Render the source slides and clean the backgrounds. Group pages by background and remove foreground content with an image editing model.
  2. Classify and rebuild components. Convert simple geometry to PowerPoint shapes, slice icons and illustrations from generated asset sheets, and crop real photographs from their source pages. Reuse components that repeat across pages.
  3. Extract text, measure layout, and match fonts. OCR reads the text, and pixel measurements recover position, size, color, and line spacing. A small local model shortlists up to 10 fonts, then OpenCV compares their glyph shapes. Reruns reuse completed text extraction results when the inputs and settings match.
  4. Compose and compare each slide. Combine the backgrounds, shapes, pictures, and text, then export previews and record visible differences against the source.

Full workflow · Text extraction and font matching

🎬 Demo Video

slide.mp4

🖥️ Examples

Retail performance

Image-based slide Editable PPTX (objects selected)
Retail performance · Image-based slide Retail performance · Editable PPTX (objects selected)

Cross-region release architecture

Image-based slide Editable PPTX (objects selected)
Cross-region release architecture · Image-based slide Cross-region release architecture · Editable PPTX (objects selected)

Warehouse redesign

Image-based slide Editable PPTX (objects selected)
Warehouse redesign · Image-based slide Warehouse redesign · Editable PPTX (objects selected)

Pendulum experiment

Image-based slide Editable PPTX (objects selected)
Pendulum experiment · Image-based slide Pendulum experiment · Editable PPTX (objects selected)

Office space planning

Image-based slide Editable PPTX (objects selected)
Office space planning · Image-based slide Office space planning · Editable PPTX (objects selected)

Self-service pilot

Image-based slide Editable PPTX (objects selected)
Self-service pilot · Image-based slide Self-service pilot · Editable PPTX (objects selected)

Assembly downtime improvement

Image-based slide Editable PPTX (objects selected)
Assembly downtime improvement · Image-based slide Assembly downtime improvement · Editable PPTX (objects selected)

City archive exhibition

Image-based slide Editable PPTX (objects selected)
City archive exhibition · Image-based slide City archive exhibition · Editable PPTX (objects selected)

Image-based PPTX · Editable PPTX · All eight comparisons · Clean overview · Original basic examples

Installation and configuration

Codex

Ask Codex to install the skill from the repository, then restart Codex.

Install the Codex skill from https://github.com/CodingFeng101/slide-alchemy.git using the skill/slide-alchemy directory.

Claude Code

Clone the repository, copy the skill into the Claude Code skills directory, and restart Claude Code.

git clone https://github.com/CodingFeng101/slide-alchemy.git
cp -r slide-alchemy/skill/slide-alchemy ~/.claude/skills/

Python, image models, and fonts

Install the script dependencies from the repository root.

python -m pip install -r skill/slide-alchemy/requirements.txt

The skill uses the host's image tool when available. For hosts without an image tool, scripts/generate_image.py uses an OpenAI-compatible image API. Set SLIDE_ALCHEMY_IMAGE_API_KEY before starting the host. Set SLIDE_ALCHEMY_IMAGE_BASE_URL and SLIDE_ALCHEMY_IMAGE_MODEL for a custom gateway or model.

The OCR model ships with the dependencies, and font matching runs locally by default. If the rebuild needs a missing font, install the cached fonts using the command below, then restart PowerPoint.

python skill/slide-alchemy/scripts/font_store.py --install-fonts

Font setup and optional remote matching · Troubleshooting

📄 License

MIT

About

A Claude Code / Codex skill that converts image-based PPTX, slide screenshots and scanned decks into truly editable PowerPoint: real text boxes, native shapes, OCR-measured layout.

Topics

Resources

Code of conduct

Contributing

Stars

53 stars

Watchers

2 watching

Forks

Releases

Packages

Contributors

Languages