I wonder if a project that replaces the "chrome" of GIMP with a different UX would be viable. Imagine a reworked menu / shortcut / dialog system that controls the unchanged core. Even better, imagine UI and UX to be live-tweakable, written in Python / Lua / Guile / you name it. That would make discovering better UI layouts and better UX flows absurdly easier.
(Yes, as an Emacs user, I want more software to be like Emacs.)
Script-fu plugin experience is definitely not great, but it has the potential to customize stuff.
Script-fu however it totally limited, it cannot access files, it cannot do anything outside of Gimp in contrast to Elisp. I wonder why that is, security reasons, as a protection from malicious plugins?
Python-fu is another option, I haven't used it but i want try it at some point. When i find some simple examples of python-fu code to learn from, I want dive into it a little bit.
Not everything can be automated, but some things can, and that's the point.
An understated benefit of a consistent UI is if the user gets stuck and searches how to do XYZ, often an LLM or search engine will give an accurate answer as it's been answered before in forums etc. But if the UI changes every few months, there's often no answer.
The UI default shipped should not materially (if any) change every few months. But power users should be able to tweak the UI and UX even more, and publish their tweaks. Some sets of tweaks might become popular among other power users, and the best finds could find their way upstream.
What I strive to achieve is speeding up the process of UI evolution, of finding better approaches to UI and UX. This may be enabled by an ability to tweak UI/UX without recompiling the whole thing, and by not having to write the UI/UX in a low-level footgun-ridden language which is C. Even now GIMP allows for quite a bit of customization within its built-in UX.
The untweaked, vanilla experience should be good enough, stable, and the norm for non-advanced users, very much like the current UX.
For a battle-tested example, look at MS Office. You can tweak its UI in rather drastic ways; I've seen VBA apps that make Excel barely recognizable, while harnessing its power. But most users never ever alter a single button on a single toolbar, and are fine following video guides showing where to click.
More generally, and just as a tangential sentiment, I don’t understand the last decade trends of dumbing down apps and removing (/not adding) features because X is enough. Back in the day we had full dialogs called by a shortcut and having searches, menu builders, etc. Nowadays everyone tries to sell only a shortcut, or only a filter, or only nothing. But why-y? Humans have multiple mechanisms of memory and orientation (and these may heavily vary across population) and people are throwing it away deliberately.
It’s all pretty antiquated and very 90s-00s level in terms of capabilities. Not even talking AI more talking text editing and non-destructive processes and GPU acceleration.
I know 3.0 aims to address some of this but it’s too little too late and you wouldn’t get much benefit being downstream from a low output team.
It just isn't the foundation you'd build on if you valued these things, they only added non-destructive recently it's not like it was core from the start, there's many other projects you could build upon that have this at their core.
Krita or honestly even Blenders shader graph would be a better and more modern starting point if that's what you wanted and you'd have a more active base where you're going to get value downstream in optimizations and more filters on a timeline that isn't multiple decades.
I also think that the effort to create such experiments should be made lower: supported, reasonably future-proof code that runs within the host application, instead of a laboriously maintained fork.