edit-image-text

name	edit-image-text
description	Edit baked-in text labels on the project's PNG illustrations (the stick-figure role cards in Bilder/, diagrams, etc.) while keeping the artwork untouched. Use when asked to rename a figure ("replace Walter with Kerstin"), fix a typo in a label, or change any text rendered into a raster image.

Editing text baked into PNG images

The illustrations in Bilder/ are raster PNGs with their labels (names like Walter, captions like Steuergerät) drawn into the pixels. There is no source SVG. To change a label you must paint over the old text and re-draw the new text in a matching font — the rest of the drawing stays as-is.

Tooling here: Python 3 + Pillow (PIL). No numpy — use pure PIL Image.load() pixel access. System fonts live in C:/Windows/Fonts/.

The workflow (proven on Walter → Kerstin)

1. Look at the image (Read tool) to see the text and where it sits.

2. Find the text's bounding box and color. Scan rows for dark, opaque pixels in the region where the text is. A blank-row gap separates the artwork from a bottom label, which gives you a clean y-range. Record the bbox, the dominant text color (e.g. (43,43,43)), and the horizontal center.

3. Identify the font. Render candidate bold fonts at a size whose cap-height matches the text height, then overlay the candidate in red over the original and Read the crop to compare letterforms. For this project's role cards the match is Calibri Bold (calibrib.ttf) at size 54, color (43,43,43), on a white background. Reuse that unless the image clearly differs.

4. Replace. Paint a white rectangle over the old-text bbox (a few px margin), then draw the new text centered on the same horizontal center and baseline.

5. Verify by reading the output PNG. The artwork must be unchanged; only the label differs.

6. Update references & rebuild PDFs. If you renamed the file, grep for the old name across *.md, update <img src=...> paths and prose, delete the old PNG, then regenerate any affected PDFs with the md-to-pdf skill (Berufsvorstellung Systems Engineer.pdf, Diplom.pdf).

Helper script

replace_label.py does steps 2–5 for a bottom-centered label and prints the detected bbox so you can sanity-check it:

python ".claude\skills\edit-image-text\replace_label.py" "Bilder\Walter.png" "Kerstin" "Bilder\Kerstin.png"

Arguments: <source.png> <new-text> <output.png>. It auto-detects the label bbox (the lowest contiguous block of dark text below a blank-row gap), erases it, and redraws with Calibri Bold matched to the original height/center/color. Always Read the output afterwards — auto-detection can misfire on images whose layout differs from the role cards (e.g. inline diagram labels). For those, fall back to the manual steps above with explicit coordinates.

Gotchas

• numpy is not installed — don't import it.
• Pillow's getbbox() returns (l,t,r,b); subtract the top offset when positioning so the baseline lands where you want.
• Match the existing color exactly ((43,43,43), not pure black) or the new label will look slightly off next to the artwork.