And whether it’s copyright infringement or not is of little interest to me since I view copyrights as being inherently harmful.

But from that perspective it would be much easier and more effective to simply use an LLM to make a crack for pirating the original software than to try to reimplement the whole thing. Or, to actually decompile and recompile it to get something (pirate-)source-available with actual feature parity. (People are also doing both of those things with various proprietary software, of course.)

However, since the implementation is using a different language, that does make the implementation itself a clean room one.

That isn't what "clean room" means in this context. It is widely accepted legal doctrine that direct ports of software from one language to another constitute derivative works under copyright law.

This project's claim to be "open source" would be false if it is actually an automated port of the output of a decompiler of proprietary programs; that would make it warez.

And it still takes time to add features using LLMs. It’s not magic, and I’m more amazed how much they managed to add already rather than it not having parity yet.

Indeed. But if they haven't gotten around to adding more than the most basic features in the Levels adjustment window (probably one of the most commonly used features of Photoshop) I assume they're missing a lot of other stuff too.

It does appear to work with PSDs fairly well, but again I don’t have really complex ones because I use the tool in an amateur way to begin with. [...] So, the feature parity isn’t really the stumbling point for me, it’s the UX.

OK, I'm taking that to mean that you haven't actually identified any specific features (or PSD compatibilities) that PhotoCraft has over GIMP besides the UX? It seems to me like using an LLM to improve the UX of GIMP, on top of all of its existing (presumably far greater) set of features, would make a lot more sense.

But this is something that’ll sort itself naturally over the coming months. If the developer stays active and continues working on it, then it’ll keep getting better. If they abandon it and move on then it’ll either die or somebody else picks it up. But it is in a fairly usable state as is from my experience, and I’d argue a lot of the hard work has already been done. So, we’ll have to just wait and see what happens with it.

I concur, in six months or a year I can either eat my words or say "I told you so" :)

And if we get to the point where all this stuff can be done using local models, then we really could see Stallman’s vision for GNU become a reality. Regular people, who aren’t developers, could use fully open source stacks, and they could modify software to work exactly the way they want. If we’re lucky this might kill closed source software entirely in the long run.

I wouldn't hold your breath, but perhaps with a lot more advances and if/when local models can perform coding tasks at reasonable speeds on affordable hardware... maybe? Meanwhile right now we're drowning in new 1M LOC slop projects overpromising and underdelivering.