I agree with you on UX. In terms of contributing to these projects, most maintainers are still opposed to LLM contributions, perhaps not for bad reasons considering they're interested in the long term maintainability not only of the software, but of a community of contributors knowledgeable enough to maintain it.
Regarding Photocraft, I've been following it with interest. My impression is that it comes off initially as an incredible accomplishment, 1:1 with Photoshop, however extended usage reveals more and more flaws until it's determined the software isn't really usable yet https://www.youtube.com/watch?v=BGShygDux9o
This is my general impression as well with overall excitement around LLM assisted reverse engineering: very easy to get something that looks like a slam dunk, until you actually try to use it. Then begins the long, long, long slog of the work needed to make the software usability and parity, which apparently can't be done by LLM agent swarms yet.
Photocraft may not be ready as a drop-in Photoshop replacement for professionals yet, and I didn't say it was. I said it was better than GIMP 3 (and all other free image editors, for that matter). I've been using pirated Photoshop for random image edits for most of my life, and Photocraft does exactly what my muscle memory expects Photoshop to do.
The video you linked is also three weeks old, which is an eternity in the scope of this project, since each week hundreds of PRs are merged, and several versions are released. At the speed they are going and with the amount of attention and usage they're getting, I expect they'll manage parity in months, but even in the current state, it's still useful enough.
The person testing it also seems to be an artist, so their workflow demands can be significantly different from someone doing image editing. Most artists I know use more specialized software like Clip Studio Paint over Photoshop anyway.
I think to stay competitive, open source software needs to adopt AI-first workflows that mostly automate quality control and review, rather than just allow LLM contributions. Allowing LLM contributions without sufficient infrastructure will just drown maintainers in slop.
> however extended usage reveals more and more flaws until it's determined the software isn't really usable yet
> Then begins the long, long, long slog of the work needed to make the software usability and parity, which apparently can't be done by LLM agent swarms yet.
We think it can be. This is just Pareto principle. We're hill climbing the finer curves.
Regarding Photocraft, I've been following it with interest. My impression is that it comes off initially as an incredible accomplishment, 1:1 with Photoshop, however extended usage reveals more and more flaws until it's determined the software isn't really usable yet https://www.youtube.com/watch?v=BGShygDux9o
This is my general impression as well with overall excitement around LLM assisted reverse engineering: very easy to get something that looks like a slam dunk, until you actually try to use it. Then begins the long, long, long slog of the work needed to make the software usability and parity, which apparently can't be done by LLM agent swarms yet.