Image to Editable PPT Skill

Image to Editable PPT Skill is an open-source Codex skill from ningzimu that rebuilds slide images, multi-page PDFs, and image-based PPTX files as editable PowerPoint presentations. Readable text becomes native text boxes, simple geometry becomes PowerPoint shapes, and complex visuals are kept as separate image assets with recorded provenance.

What It Is

The skill turns screenshot-style or image-style slides into .pptx files whose text, simple shapes, and visual assets can be adjusted separately. It does not generate a new deck from an article or outline; the same author's codex-ppt-skill covers that case. Output is always a PowerPoint file, with one output page per input page in the original order.

How the Conversion Works

The input is first normalized into per-page tasks. A single page can be rebuilt locally by the main agent, while multi-page inputs are dispatched to page workers and processed in parallel up to a configured concurrency limit. Each page gets a manifest describing its text, shapes, and assets. The final deck is then assembled from the recorded manifests and validated.

• Text size and position are driven by measurements: an OCR step (Baidu PaddleOCR-VL, when a token is configured) supplies box coordinates, font sizes, and size groupings. Without a token, it falls back to a built-in offline detector with lower text fidelity.
• Image generation and editing for background completion and icon extraction use Codex's built-in image tool first, then fall back to the editppt image CLI via Codex OAuth or an OpenAI-compatible API.
• Regular row-and-column tables can be rebuilt as native PowerPoint tables, and native curves, dashed lines, and arrowheads are supported.
• Notes from PPTX inputs are copied unchanged to the matching output pages.

Setup Path

Install by asking the agent to install the skill from the GitHub repository. In Codex it is invoked as $image-to-editable-ppt with images, PDFs, or PPTX files attached or referenced by path. The docs recommend Codex's full-access permission mode because the run includes OCR, image generation, file writes, and sub-agent dispatch. Fallback API settings are stored in a user-level config file.

Tradeoffs to Know

The README states that the workflow is token-intensive, possibly two to three times the cost of generating an image-based deck, and that single pages can take over ten minutes. It does not guarantee a 100% faithful copy of the original page, and photos, illustrations, and textures can only be moved as image assets, not edited internally. It is not guaranteed to work well with models below gpt-5.5. Native tables do not support diagonal headers or embedded cell images.