unflatten-ppt

name

unflatten-ppt

description

Convert PNG screenshots, raster diagrams, generated figure drafts, or reference images into editable PowerPoint slides. Use when the user asks to turn a PNG/image into PPT, recreate a figure as editable PPT elements, make every object editable, export screenshots for comparison, iterate against a reference image, or build a slide from a raster technical diagram instead of embedding the image as a flat bitmap.

PNG to Editable PPT

This skill rebuilds a raster reference image as an editable PowerPoint slide using native PPT objects: text boxes, shapes, connectors, arrows, tables, and grouped layout elements. Do not satisfy the request by pasting the PNG as a single background image unless the user explicitly asks for a non-editable raster slide.

Core Workflow

1. Inspect the source PNG dimensions and content.

   • Identify the page size, major regions, text blocks, shapes, arrows, colors, and repeated structures.
   • Use the PNG pixel coordinate system as the first layout coordinate system.

2. Generate an editable PPT reconstruction.

   • Prefer python-pptx for deterministic construction.
   • Set the slide size to match the PNG aspect ratio.
   • Map pixels to EMUs with a simple helper such as EMU_PER_PX = 9144 when using 96 dpi coordinates.
   • Recreate all important elements as editable PowerPoint objects.
   • Keep source-generation code whenever the slide is complex, so future iterations can be regenerated instead of hand-edited.

3. Export the slide to PNG and compare.

   • Use scripts/export_ppt_slide.ps1 on Windows when PowerPoint is installed.
   • Use scripts/compare_pngs.py to create a side-by-side comparison and basic image-difference metrics.
   • If PowerPoint export is unavailable, use any available office renderer, then state the limitation.

4. Iterate.

   • Compare the export against the reference image.
   • Fix layout first: canvas size, outer frame, large blocks, and section boundaries.
   • Then fix text: font sizes, wrapping, alignment, labels, and clipped text.
   • Then fix connectors: arrow direction, endpoints, line thickness, dashed lines, and routing.
   • Repeat until there are no obvious structural errors.

5. Deliver artifacts.

   • Put the editable .pptx in the requested output directory.
   • Also keep the latest exported PNG and a side-by-side comparison PNG when the user asks for screenshot comparison or iterative refinement.
   • Report exact file paths and any remaining known deviations.

Reconstruction Rules

• Editable means native PPT objects, not one flattened image.
• Preserve the diagram semantics over pixel-perfect ornamentation.
• Use the source image as a visual reference, but do not trust OCR blindly; manually verify labels.
• Use conservative fonts such as Arial/Calibri unless the source clearly needs another font.
• For technical figures, preserve identifiers exactly: underscores, parentheses, operators, array indices, and capitalization.
• Use native table/grid cells for register files, matrices, queues, and repeated boxes.
• Use native connectors for arrows whenever possible; if python-pptx cannot express a required arrowhead directly, patch the OOXML or document the limitation.
• Keep all generated filenames descriptive and versioned, for example figure_editable.pptx, figure_export_r1.png, figure_compare_r1.png.

Useful Scripts

• scripts/export_ppt_slide.ps1: export one PowerPoint slide to PNG through local PowerPoint COM automation.
• scripts/compare_pngs.py: create a side-by-side comparison PNG and print RMS/mean-difference metrics.

Read references/quality-checklist.md before final delivery.

Typical User Requests

• "Convert this PNG into an editable PPT."
• "Make every element editable; do not paste the whole image."
• "After each version, export a screenshot, compare it with the reference, and iterate."
• "Rebuild this paper figure, architecture diagram, or flowchart from an image into PowerPoint."
• "Make this raster diagram into an editable PowerPoint slide."