Image to Editable PPT takes screenshots, scanned slides, PDFs and image-based decks and rebuilds each page as a .pptx whose text, simple shapes and tables are native PowerPoint objects. Complex visuals such as photos and illustrations stay as separate image assets with a recorded source.
It runs through an editppt command-line runtime that tracks every page from prepare to dispatch, record and finalize. Multi-page inputs are rebuilt by page workers in parallel, a single page by the main agent, and the final deck is checked for structure, slide count and media links.
What You Get
- A .pptx with one slide per input page, in the original page order
- Text restored as native text boxes, with font sizes grouped by level from measured values
- Simple geometry, lines, curves and ordinary tables rebuilt as PowerPoint objects
- Photos and complex visuals split into separate movable image assets with provenance records
- Speaker notes copied unchanged from an input .pptx, plus a validation report for the deck
Common Use Cases
- Make a deck that was exported as images editable again
- Rebuild a slide from a screenshot with adjustable text and shapes
- Convert a multi-page PDF of slides into a single editable .pptx
- Reconstruct an image-based PPTX while keeping its speaker notes
What Makes It Unique
Rules live in one place each, and run state advances only through editppt commands, so a long conversion can resume from files. A page is recorded only if its manifest rebuilds the page and its validation passes, and a full-slide picture with text laid over it is rejected as a fallback. Text size and position come from OCR measurements, so the agent works from numbers for every text box.
Under The Hood
The agent installs and calls the editppt command-line tool, a Python runtime shipped in the skill’s cli folder, and runs a prompt-builder script for page workers. It reads your input files and writes a job folder with page images, manifests, previews, assets and the final deck. When a PaddleOCR token is configured it uploads page images to Baidu’s OCR service, and it asks you once for that token. Image generation and editing for background repair and asset extraction go to the agent’s built-in image tool first, then to a Codex login or an OpenAI-compatible API whose key is stored in a config file in your home folder. Multi-page runs need an agent that can spawn sub-agents, and long runs spend a lot of tokens.