1. Can you post the picture with the mask "before" you piush it?
   
2. There are globally three ways to represent channels in Gimp:
   • As a value in the [0.0 ... 1.0] interval (Pixel in the Pointer and Sample points dialogs). This is mathematically how thing work under the hood in 2.10)
     
   • As a percent [0% ... 100%] (RGB(%) in said dialogs), so 100% is the same as 1.0 in Pixel.
     
   • As a 8-bit channel value in the [0 ... 255]  range (RGB(0..255) in said dialogs) so 255 in the same as 100% in RGB(%) or 1.0 in Pixel.
     
3. So your opacity is likely expressed in percent...
   
   To add to the fun the Pixel value is a linear scale, while the RGB values are "perceptual", so a "middle" gray (gray that has the same overall luminosity as a dense checkerboard) is .50 in Pixel, but 73% (RGB(%)) and 186(RGB(0..255))
   

(12-12-2023, 11:10 PM)Ofnuts Wrote:

1. Can you post the picture with the mask "before" you piush it?
   

• If I use your "Before" image, and threshold the mask to 0.06 I get essentially the same mask as in your "impossible colors" picture, but the image looks good.
  
• If I copy/paste that new mask to the mask of the bad picture, the picture is fixed
  
• If I copy the mask of the bad picture and the mask of the good picture as layers in a new picture and compare them... more weirdness ensues. The masks look identical, so putting the top one in Difference mode should give a black image.... and it doesn't. And when I use the Pointer dialog to look at the mask values, these values are completely out of range. The values that should all be 0. <= x <= 1.0 (as they are in the mask if the good picture) are in the thousands range! Also, they appear quite random, and two adjacent pixels have very different values.
  
• So my guess is that due to the out of range values in the mask, the code that computes the final color gets out of range colors and what is displayed is a bit random.
  

(12-15-2023, 07:27 PM)Grobe Wrote: I found the original image of the generator. The problem is I don't manage to reproduce the issue by now.

However, I do have a theory about how this came to be. In order to get to a layer mask that resembles a grayscale copy of the original image, I use selection first, and then create the layer mask from selection. Therefore I suspect that I've forgotten to "select none" before further editing.

This is what I do have by now, not sure if it is useful by now.

Ok - cannot upload new atachment, the xcf files are too large.

(12-15-2023, 08:47 PM)Ofnuts Wrote:

(12-15-2023, 07:27 PM)Grobe Wrote: I found the original image of the generator. The problem is I don't manage to reproduce the issue by now.

However, I do have a theory about how this came to be. In order to get to a layer mask that resembles a grayscale copy of the original image, I use selection first, and then create the layer mask from selection. Therefore I suspect that I've forgotten to "select none" before further editing.

This is what I do have by now, not sure if it is useful by now.

Ok - cannot upload new atachment, the xcf files are too large.

You can save the XCF with advanced compression (shrinks to around 1/3 of initial size).. The limit on XCF is 2MB and with the compression this should be enough?