ningzimu/image-to-editable-ppt-skill

Overview

A skill for converting images, PDFs, and image PPTs into editable PowerPoint files. It first normalizes input into page-by-page tasks, then rebuilds it into .pptx: readable text is restored as much as possible to native text boxes, simple geometry is restored as much as possible to PowerPoint shapes, and complex visual elements are retained as independent image assets that provide source records. It's ideal for turning screenshot-style or image-style slides into PPTs that are easier to edit, allowing text, simple shapes, and visual materials to be adjusted as separately as possible. This skill has a long runtime and automatically executes OCR, image generation/editing, file read/write, sub-agent dispatch, and long polling steps. The "request approval" mode frequently interrupts execution and may block some steps, especially in the sub-agent environment. The "Approve for me" mode is known to still intercept requests during OCR, ChatGPT image generation/editing, or third-party API call stages, requiring manual approval; If you are not near a computer, the conversion process will stop. During conversion, this skill automatically calls Baidu's PaddleOCR-VL interface (if tokens are configured) to correct page text boxes, font size, and font size grouping, and also calls ChatGPT's gpt-image-2 image generation/editing interface for front background separation, icon/visual material extraction, and partial image restoration. These calls are necessary steps to reconstruct an image page into an editable PPT. Currently, Skill uses a multi-agent collaborative recovery process with complex process control, not a lightweight converter. AI executes "" loops and may iterate multiple times until it considers the result close enough to the original. During this process, page workers may make very difficult moves on the pages, so overall it consumes more tokens. Restoring a 10-page PPT can consume your 5 hours of credit. The recovery time for a single PPT slide may exceed 10 minutes. Single-page or single-image input can be locally executed by the main agent according to the same page reconstruction process; Multi-page input is assigned to the page worker according to concurrent slots. A lighter approach is to directly use gpt-image-2's image editing capabilities: send the PPT image you are not satisfied with, have it modify it accordingly, and return the modified image. This skill does not directly generate a brand-new PPT from articles, reports, outlines, or ideas. If your goal is to "generate a PPT," you can use the codex-ppt-skill. For a detailed introduction to the codex-ppt and image-to-editable-ppt skills, see skill_duo_intro.pdf.

Overview (machine translated)

This page was translated automatically. Technical details may differ from the original.

Recommended Tools

Try a different keyword or remove a filter.

Install