Real Transparent Background

Quick Start

Use scripts/checkerboard_to_alpha.py for deterministic background extraction when an image already contains a visible gray/white checkerboard or similar fake transparent background. Prefer this script over AI image regeneration when the foreground must remain unchanged.

Run it with a Python that has Pillow and NumPy available:

python "/path/to/real-transparent-background-skill/scripts/checkerboard_to_alpha.py" "/path/to/image.png" --suffix=-transparent

For multiple files:

python "/path/to/real-transparent-background-skill/scripts/checkerboard_to_alpha.py" "/path/to/a.png" "/path/to/b.png" --output-dir "/path/to/transparent" --suffix=-transparent

Workflow

1. Treat the provided image path(s) as edit targets, not style references.
2. Save non-destructively by default. Use a sibling filename such as name-transparent.png unless the user explicitly asks to overwrite.
3. Use the bundled script for fake checkerboard or flat neutral backgrounds. It estimates the background colors from border pixels, flood-fills only background connected to the canvas edges, and writes an RGBA PNG.
4. For images with colored glows, halos, lightbulbs, or aura effects that should remain semi-transparent, add --soft-colored-edges. For thin pale foreground details such as whiskers or white line art, omit that flag to preserve colored details.
5. If writing outputs outside the current workspace, request filesystem approval before copying or saving there.
6. Validate the result before finishing: output mode should be RGBA; corner alpha should be 0; foreground sample points should remain 255.

Script Notes

scripts/checkerboard_to_alpha.py supports:

• one or more input files
• --output-dir for batch output
• --output for a single explicit output path
• --suffix for sibling filenames
• --overwrite only when replacement is intentional
• threshold tuning for unusual backgrounds

The script prints one JSON line per processed file with output path, image size, palette estimate, alpha counts, and corner-alpha samples. Use those values in the final response when reporting validation.

When To Avoid The Script

Use a true segmentation or model-native transparency workflow instead when the background is not connected to image borders, is a complex scene, overlaps foreground colors heavily, or the task requires inventing missing foreground pixels. Ask before switching to a generative or API fallback if it could alter the subject.