Reverse Engineering PowerPoint's XML to Build a Custom Slide Generator

Making PowerPoint presentations can be deceptively challenging, especially when you need them to look polished and professional. While there are several AI-powered slide generators available, they often fall short of producing truly satisfying results. I decided to build a better one, diving deep into the 5,036-page Office Open XML documentation to understand how PowerPoint files work under the hood.

The PowerPoint Maze

To generate custom PowerPoint presentations, I needed to understand their internal structure. Microsoft’s Office Open XML standard (ECMA-376) is the key here, but it’s a behemoth at 5,036 pages. For those who haven’t delved into this yet, here’s a quick overview:

A .pptx file is essentially a zip archive containing multiple XML files and media assets. You can explore its contents by simply unzipping the file:

unzip pres.pptx

Each slide, chart, speaker note, theme, master slide, and layout has its own XML file. These files reference each other in a well-defined but complex manner. The complexity is such that PowerPoint can be Turing-complete just through shapes and animations (no scripting required)!

XML Structure of a Slide

Let’s break down the structure of a typical slide:

• Slide XML Files: Each slide has an associated slideX.xml file, where X is the slide number. This file contains all the elements on the slide.
• Media References: Images and other media are stored in the media directory. The XML files reference these assets using relationship IDs (rIds).

For example, here’s a snippet of an XML subtree for a picture:

<Relationship Id="rId5" Type="http://schemas.openxmlformats.org/officeDocument/2006/relationships/image"
    Target="../media/image41.png" />

• Picture Elements: The <p:pic> tag represents the image, with child tags like <p:nvPicPr> for non-visible properties and <a:xfrm> for position transformations.
• Image Path: The actual path to the image is specified in another XML file, slideX.xml.rel, using the relationship ID:

<a:blip r:embed="rId5" />

Generating the Deck

Once you understand where to place the right XML tags, generating a custom slide deck becomes more manageable. Here’s a simplified example of how to set the geometry and add adjustments for a rounded rectangle shape:

<a:prstGeom prst="roundRect">
    <a:avLst>
        <a:gd name="adj" fmla="val 2430" />
    </a:avLst>
</a:prstGeom>

Challenges and Solutions

Building the slide generator wasn’t without its challenges:

• Complex XML Structure: The interdependencies between XML files required careful handling to ensure everything linked correctly.
• Image Formatting: Achieving precise formatting, such as rounded corners, involved digging into specific tags and attributes.

Despite these hurdles, the process was rewarding. By the end of three weeks, I had a functional slide generator that could produce high-quality presentations tailored to our needs at Listen, where we conduct user interviews and need to present customer research insights effectively.

Conclusion

Reverse engineering PowerPoint’s XML structure allowed me to build a custom slide generator that meets our specific requirements. While it was a complex journey, the result is a tool that produces polished, professional slides with minimal effort.