notebooklm-to-editable-pptx

name	notebooklm-to-editable-pptx
description	Convert a NotebookLM generated image-based PPTX into a fully editable PowerPoint presentation (PPTX) using native macOS Vision OCR.
allowed-tools	Bash, Read, Write, Glob, Grep

NotebookLM to Editable PPTX Conversion

Overview

NotebookLM currently generates PowerPoint (.pptx) presentations where every slide is a flattened, uneditable image. When you need to convert a NotebookLM export into an editable .pptx file, leverage this skill.

This workflow uses a native macOS Swift script (Vision framework) for highly accurate OCR on the images, followed by programmatic reconstruction of the presentation using python-pptx.

Prerequisites Ensure

1. python3 and pip are installed.
2. The python module python-pptx is installed:

   pip3 install python-pptx
   

3. A macOS environment (the Swift OCR script relies on Apple's Vision framework).

Workflow

Follow these steps precisely:

1. Extract Images from the NotebookLM PPTX

Use the extract_images.py script provided in this skill's scripts/ folder to pull the PNG images out of the uneditable presentation.

# Locate the script first
export SKILL_DIR=$(dirname $(find ~ -name "extract_images.py" | grep "notebooklm-to-editable-pptx" | head -n 1))

# Extract images to a temporary directory
mkdir -p /tmp/pptx_images
python3 "$SKILL_DIR/extract_images.py" "path/to/NotebookLM_Export.pptx" /tmp/pptx_images

2. OCR the Extracted Images

Use the ocr.swift script to perform Optical Character Recognition (OCR) on all the extracted PNG files. Save the output to a text file for analysis.

# Clear the text file first
> /tmp/pptx_extracted_text.txt

# Run OCR on all extracted slides (Slide count must match the extraction output)
for i in /tmp/pptx_images/*.png; do
  echo "--- $i ---" >> /tmp/pptx_extracted_text.txt
  swift "$SKILL_DIR/ocr.swift" "$i" >> /tmp/pptx_extracted_text.txt
done

3. Analyze the Extracted Text

Read the /tmp/pptx_extracted_text.txt file and carefully parse the content for each slide:

• Distinguish between Title text, Body points, and Table Data.
• Identify the logical structure.

4. Generate the Editable PPTX

Write a custom Python script for the specific presentation using python-pptx. Map the extracted OCR text into real, editable python-pptx Layouts (such as Title Slides, Title and Content, Tables, etc.).

Guidelines for the generation script:

• Use prs = Presentation()
• prs.slide_layouts[0] is for Title Slides.
• prs.slide_layouts[1] is for Title and Content (bullet points).
• For tables, use prs.slide_layouts[5] (Title Only) and draw a table via slide.shapes.add_table(rows, cols, left, top, width, height).
• Reconstruct the slide content logically based on your OCR analysis.

# Example generic execution
python3 custom_pptx_generator.py

When to Use

This skill applies whenever a user uploads a PowerPoint (.pptx) generated by NotebookLM (or any presentation where the slides are purely flattened images) and explicitly requests to make the text "editable", "convert to a real PPTX", or "recreate the layout".