ningzimu/image-to-editable-ppt-skill

What it solves

This project provides a way to convert non-editable images, PDFs, and image-based PowerPoint presentations into fully editable .pptx files. It solves the problem of having to manually recreate slides when you only have screenshots or flat image files, allowing users to edit text, move simple shapes, and manage visual assets independently.

How it works

The system uses a multi-agent collaboration process to reconstruct the slides. It follows a "reconstruct $\rightarrow$ self-check $\rightarrow$ correct" loop to ensure the output is as close to the original as possible.

Key technical steps include:

• Normalization: Inputs are converted into page-by-page tasks.
• Text Recovery: It uses OCR (specifically PaddleOCR-VL via a third-party token) to measure exact coordinates, font sizes, and groupings to restore native text boxes.
• Visual Reconstruction: Simple geometric shapes are restored as PowerPoint shapes, while complex visual elements are extracted as independent image assets.
• ** uma image backend**: It prioritizes built-in image generation tools (like Codex image_gen.imagegen) and falls back to OpenAI-compatible APIs for image editing and asset extraction.
• Parallel Processing: For multi-page documents, the main agent dispatches tasks to "page workers" (sub-agents) to process pages concurrently.

Who it’s for

Users who need to transform static slide images or PDFs into editable presentations for further modification, specifically those who want to preserve the layout and text of a visual design while regaining the ability to edit the content.

Highlights

• Broad Input Support: Handles single images, multiple images, multi-page PDFs, and image-based .pptx files.
• Measurement-Driven Text: Uses OCR to ensure text size and position are based on actual measurements rather than visual estimation.
• Hybrid Reconstruction Strategy: Combines editable text, native PPT shapes, and independent image assets for complex pages.
• Multi-Agent Workflow: Employs a sophisticated iterative loop of reconstruction and validation to improve accuracy.
• Note Preservation: Automatically copies original page notes from .pptx inputs to the final output.