
It's a known issue. The image is converted into a color palette, but the script sometimes blends colors into the palette. These colors are then assigned matching names.
The process should be improved: instead of creating palettes this way, all colors should be properly quantified and named only if they are sufficiently present in the image.