guopenglin/image2ppt

DeckWeaver is a Python tool that takes images of slides, such as the screenshots that GPT or Gemini might produce, and rebuilds them as editable PowerPoint files. The README is mostly in Chinese, with an English version available in a separate file. Almost all text on the input images is recreated as real PPT text boxes, and the icons, logos, pictures, and decorative elements are extracted as separate image objects that you can move or replace inside PowerPoint. The project pitches two main use cases. The first is using it as a Skill or tool for an agent like Codex or Claude Code: you clone the repo into your skill directory, point the agent at a folder or a single image, and the agent runs the conversion through a bootstrap script. The second is using it as a standalone command line tool for people who do not want to spend tokens on a large model, do not need perfect text accuracy, and want to process many images in batch. The README says the heavy work runs locally. OCR, image segmentation, PPTX generation, and a preview check all happen on your own machine, so no cloud API key is required. It accepts common bitmap formats such as PNG, JPG, WebP, BMP, and TIFF. A bootstrap script installs Python dependencies, a local OCR stack, and LibreOffice or Poppler for preview rendering. macOS and common Linux distributions are supported directly, while Windows users are pointed at the requirements file for manual setup. A convert script is the one shot entry point, and there are flags to limit pages, skip rendering, try to detect tables, or export icon decisions for human review. The results land in an output_project folder that includes the final pptx, a structure check report, preview images, OCR files, layout JSON, and extracted assets. The author notes that complex charts are kept as movable image objects rather than rebuilt as native PPT charts. The license section says personal use, copying, and modification are free, but copies and derivatives must keep the original credit, license, and repo link. Commercial use, redistribution, SaaS integration, and internal production deployments require contacting the author for a paid commercial license at the email in the README.