Problems with 'Split Layer' function

1 Obviously different colors are recognized as the same ones. (It almost feels like the color detection is based on value and not the color itself… but I wouldn’t know.)

2 If I lower the fuziness to avoid the problem 1, the detection becomes too precise and hundreds of additional tiny layers are created.

3 The detected edges are very rough at any fuziness level, which makes this function basically unusable.

Your colour wheel has many .jpeg artifacts at the boundaries which affect the local colour. Ideally, you’d have a good quality .png source.

You will have a hard time getting it to work with any image with such a poor quality.

If the source image is a photo, or a painting, there should always be many ‘artifacts’ over the image anyway even if it has a good resolution quality. Because that kind of niose is just a natural element.

If the function is supposed to be only used on ultra-clean-vector-like images, I get it. But if not, it has the problems that I mentioned.

More or less it happens with any images except ultra-clean-vector-like ones, but I get that you wouldn’t know it is because it seems you have fairly less experience on image editing and painting. :slightly_smiling_face:

Real life is ‘noisy’, definitely. Mathematical functions are precise and they don’t know about context or what may make sense to you. This is always the problem when you try applying these functions to real images.
You have to work with what you have and work around any problems and what the application is capable of. I’ve no idea how you could do that with this particular function.

Yes, my portfolio clearly shows that I have no Idea what I’m talking about.

One of the major use cases the feature as implemented was to separate flats ( flat colors put by artists while making comics) the quality of flats is generally higher than your source image hence it works with minor tweaks in the threshold value. So I would say as per my information it is not designed to cleanly separate images like you have shared. May be there can be improvement in it in future. That may depend on any new use case that comes up.

P.S. Please don’t make the discussion personal. Remember the rule always be civil.

1 Like

Ok. I probably am trying to get something that cannot be done, about the third one.

But I still think it’d be better for it to devide and group the areas more smartly. The purpose I want to use this function is to edit different colored areas separately, and it’s hard for me to get properly separated areas with the current system. (1,2)

If that’s also related to the noise, I don’t have anything more to say. :roll_eyes:

It’s a bit more manual work but you could use selection masks or saved selections for this. Automation has it’s limits but the new lasso tool could maybe make it less of a chore.

Not everything can be done with every feature. I think this feature is mostly a simple separation for output of Colorize Mask and GMIC’s Smart Colorization features.

In this case I would suggest using Continuos Selection Tool, with Fuzziness around 12 (can be a bit higher for more noisy colors) and Grow 1 (to not have to redo the background), select color, copy and paste, repeat for all colors.