Port Fuzzy Selection from GIMP as a mask creation tool, alternative to AI #20538
Description
Is your feature request related to a problem? Please describe.
Modern editing workflows in other image editors depend on precise, ideally pixel-perfect, object masks. AI models such as SAM2.1 are the primary solution proposed (see #20322), but the AI topic is very polarizing. Some people do not accept the current AI models on ethical grounds but still deserve pixel-perfect object masks, even when current parametric masks are insufficient.
Describe the solution you'd like
GIMP already has a non-AI tool for this exact purpose: Fuzzy Selection. And it performs similarly well to SAM 2.1, or in many cases (e.g., bird feathers) significantly better, with just a bit more clicking required. Hence, here is the request for porting it.
Demo in GIMP: https://discuss.pixls.us/t/for-something-different-sky-selection/56371/12
Alternatives
Continue creating non-AI masks with GIMP as an external tool, or surrender to the AI process.
Disclaimer: I am a user of AI masks with ART, but at this point, I sometimes prefer GIMP's fuzzy selection on the grounds of better quality of the resulting mask, and I would like AI opponents to have a viable alternative to AI masks anyhow.
Additional context
The AI mask merge request already contains the infrastructure for selecting a set of points that define the input for producing a mask, and that's almost exactly what Fuzzy Selection needs. Almost, because there are extra parameters associated with each point, like the "Threshold" slider and the "Select by: [Composite, Red, Green, Blue, ...]" choice box. In addition, it is often required to remove holes from the resulting mask, so an interface for doing so should also be provided. It may also be a good idea to offer an option to run the fuzzy selection tool on a slightly blurred image so that the fine texture is ignored.
@andriiryzhkov @TurboGit To be clear, this does not nullify your work on adding AI-based masking to DT. We need both ways of creating object-based masks.