The auto generation type selection is broken since the commit reference range is added.
The old conditioning was checking whether the captured image and masks are both solid color (i.e. all pixels are same color), but obviously the mask of the expanded area will be black while the selected area will be white, so the condition for txt2img is only satisfied when selecting the whole canvas, and the whole canvas is solid color.
After using the the feature for several weeks, I found that it is pretty annoying an unpredictable, as a single pixel in the selection can break the logic, leading to wrong generation type, thus I propose removing the Auto generation type feature.
The auto generation type selection is broken since the commit reference range is added.
The old conditioning was checking whether the captured image and masks are both solid color (i.e. all pixels are same color), but obviously the mask of the expanded area will be black while the selected area will be white, so the condition for txt2img is only satisfied when selecting the whole canvas, and the whole canvas is solid color.
After using the the feature for several weeks, I found that it is pretty annoying an unpredictable, as a single pixel in the selection can break the logic, leading to wrong generation type, thus I propose removing the
Auto
generation type feature.