Link: https://github.com/storytold/photocraft

Even if it works despite being AI slop that's been developed in a mere 4 days (?) and somehow has all these contributors, I'm not sure I agree with the idea.

It's cool that they went with Rust for it, but I would have much preferred if all that effort went towards improving GIMP and existing open formats.

If it works then I guess it's the same end result at the end of the day, but it would suck if GIMP doesn't end up being the one to put a nail on Photoshop's coffin after all these years of serious development by humans.

Edit: I already see the kneejerk dislikes coming, I'm just interested in starting a conversation about the idea of it, I also dislike the fact that it's AI slop as I already implied. I'd appreciate joining the conversation instead of dislike bombing with no argument.

It's an interesting exercise. If agentic AI is capable of easily recreating commercial software, I wonder if that'll make some software companies change their stance on it.

Not interested at all in using it though. I'm sure it'll be buggy as hell and I much prefer using software developed with actual care and purpose.

What effort? Having an LLM churning on a problem isn’t effort.

Not really a clean room implementation. Source has been leaked. Plus GIMP and other FOSS tools that implement the same functionality are in the training data. Very likely this "clean room" implementation is just laundering code.

I will be extremely surprised if this comes anywhere near gimp. AI is not at the stage were is can create amazing software all by itself. It will likely be buggy and/or feature incomplete.

I will never trust something genetated by llm and barely reviewed by human. Ai agents already showed they are willing to cheat, lie and then cover it up. No no no.

God I wish one of these megacorps would bite on suing one of these projects, if only so there is precedent of punishment for AI code-theft. The fact that they won’t touch them for fear of damaging their golden goose means it’s de-facto fair use.

Clean room reversing is just laundering code, always has been.

Rewriting code into English/GGUF/Klingon/whatever, and then getting someone else to rewrite it back into code is just laundering it. Its been legally accepted for a while, and many projects we take for granted today started that way.

Whether this case is "clean" enough is up to the courts ultimately, but its laundering either way.

Arguing such thing is "open source" is misleading as 1/ contributors can hardly call themselves "authors" of this code 2/ nobody can demonstrate that the plagiarizing engine they used did not plagiarize proprietary source code or source code published under an incompatible license.

as in "you cannot claim a room is clean if you have never been in it, haven't built it, and cannot even know what's in it"

Same, and I'd love for copyright to die, except I need capitalism to die first

If the authors used the same method to improve GIMP, they'd be introducing a bunch of slop code to GIMP. This code is not going to be of GIMP quality.

I also doubt this will hold a candle to GIMP for the aforementioned reasons. It might appeal to a small audience, but I do not expect it to dethrone GIMP any time soon.

even the people over on hacker news concede that they are unusably buggy so.. i think vibe coded slop is vibe coded slop - use gimp

Not to mention the hoard of software patents Adobe will unleash their horde of lawyers with if this becomes anything more than a bad joke. Although that would make for an interesting precedent.

I find clean room implementations to theoretically be cool, but there's a big area for bugs because they're based on feature parity, to my understanding. I would love for a 1:1 recreation of software like Photoshop, but unless there's something that can also be done to iron over bugs, I doubt the implementation would be good.

the room can be spotless, but if your reagents are all tainted you'll only ever get slop plagiarized from other projects. but sure, enjoy your "clean room re-implementation of Photoshop" XD

It's AI slop. It's not worth a discussion.

Use Krita or Gimp.

In this case the clean room is probably "We told Claude to write it and we don't care if Claude has a bunch of leaked Adobe source code in it's training data."

well fortunately for them, moth supposes, that they don't need amazing software. they just need enough to beat Gimp, which has sucked since moth was a larva. it's a rather low bar to beat.

anything to avoid having to use Gimp, moth supposes. moth has been using Gimp since moth was a larva, and it's always been one of the worst experiences in open source software.

Either Adobe wins which maybe will shut up vibe coders or vibe coders win which maybe will shut up Adobe. Either way win win

From what I gather right now it’s the UI of Photoshop with capabilities of paint.

As I have said many times: license washing works both ways, this is the end of proprietary software as we know it, no more would I have a need to install a sketchy driver inside a windows VM just to use some crap USB microscope

Just more bullshit hype. This project will go nowhere.

Imagine being this mad at gimp. lmao

Do they even say which version of Photoshop they're mimicking?

I have to admit, "AI copyright laundering Mexican standoff" isn't an angle I had considered.

I think this is a nice test to see, if license washing machines can defeat adobes lawyers.

moth isn't mad at gimp for stagnant mediocrity any more than moth is mad at league of legends for it.

From their own documentation:

opening and re-saving a document renders the same for 307 of the 309 files in the psd-tools test set and 169 of 170 in our mixed ag-psd/psd-tools set

Just what I want in my editing software, the assurance that my saved files will mostly render properly when reopened in the same program...

This is exactly what will sink any attempts to do similar things, it's not the broad functionality of the software, it's all the billions of edge cases.

The LLMs will be able to get all the general requirements mostly right, and that's where it will stop. The subtle errors and bugs will pile up steadily as the codebase grows too large and complex for even frontier models to accurately summarize, and the people who vibed it to begin with don't actually understand what is happening, so they can't troubleshoot effectively.

Automated bug reports will pile up, agents will be dispatched in greater numbers, leading to cost issues and further complicating and compounding the errors in a bloated and convoluted codebase. The leads of the project will get burned out and abandon the project, leaving a giant, rotting, husk behind. Software that was forever, "almost working properly."

Honestly it's one of the few scenarios I did consider at the start. On the one hand, your megacorps like Disney (for example) absolutely want to replace human artists with slopbots. On the other hand, Disney does not want Joe Rando to be able to type "make me a Spiderman comic" into the slopbot.

AI slop.

On the one hand, I am opposed to AI slop.

But on the other hand, I am greatly in favour of anything that makes life even slightly more difficult for Adobe.

This is a tricky one.

It's really funny going back and forth between lemmy and HN to compare the reactions

It is built by AI agents, for AI agents. I guess it's interesting seeing software built from the ground up with hooks built in for every single action to be controlled by an AI agent, in that sense you could direct your own AI agent to use this to create/edit images. But even then I wouldn't trust this thing to edit and save over existing images, it is way too new and will have plenty of non-obvious bugs that would get embedded in every image you save.

And like most vibe coded AI projects this one will probably cease being maintained long-term.

If you're actually going to work with images you plan to keep long term just stick to GIMP, Krita, etc.

It's funny to see corpos get mad but honestly a project this big, spun up this quickly is almost certainly an unmaintainable mess.

I think there are good reasons to be wary of slop that don't involve animorphing into an MPAA lawyer and parroting screeds about how you shouldn't download cars or whatever.

But also fuck Adobe

It would be okay, if it wasn't Ai. I think using LLMs cannot be guaranteed to be a clean room implementation.

On the one hand, I am opposed to AI slop.

Me too. But what is Ai slop? To me it is either fully vibe coded, or partially. Either way the user of the Ai ends up not knowing exactly what is going on. That is slop. But if its very carefully used, analyzed, understood, iterated and tested, like a human code would be, then I cannot call it slop. But it doesn't take away the problems using Ai in the first place. There are inherent ethical issues, and questions if the licensing and copyright will become an issue one day.

In short, not every usage of Ai is slop. But every usage of Ai has ethical unanswered questions.

Interesting idea that probably does not deliver. Looks like their funding model is in-app AI gen in the cloud that you buy credits for. I'm always looking for an Illustrator alternative but have yet to find one. Lots of good vector drawing apps but Illustrator is so much more than that...

The app icons look like the the prompt was: draw [random animal] in style of O'REILLY programming book covers...

Looking at the PR's & git activity & the interest in this whole project, I doubt LLM powered humans couldn't keep this running if nothing unexpected -like a lawsuit- happens.

My personal experience with LLMs is that they are very good for adapting existing code to newer libraries & various other changes.

AGENTS.md in the repo, btw.

Most of these are very modular, so should be very easy to maintain. I guess we'll just have to see.

Software that was forever, “almost working properly.”

That describes how I feel about MS Word quite well.

The difference is the original source code, and similar source code that is incompatibly licensed, was in the LLM training data, making it not a clean room.

I would have much preferred if all that effort went towards improving GIMP and existing open formats.

Well... if it was a 4-day vibe-coding project, then that's not a lot of effort, and I don't think GIMP wants to deal with that scale of AI contributions anyway.

Let's grab our popcorn and watch Adobe fight AI companies on this. Because if this isn't legal, then what does that say for everything else generated by AI?

Oh yeah my coworker was walking about this. Someone spent about $2k in tokens recreating the Adobe suite on windows, macos, and linux. Crazy what AI is able to do nowadays.

I don't think I will use it but I feel like in a year or two, software written like this will be everywhere

I'm interested in where it goes in regards to legal drama. It can't be used to improve gimp or anything else really though. I don't think it actually meets the definition of clean room which would cause issues if anyone sought legal action, and many foss projects don't take AI contributions at all anyway.

They aren't saying that opening and saving their own files corrupts. They are saying their psd saving is 99% the same as adobe's.

Thats not a difference. Other clean room projects, for example ReactOS, have source projects that are incompatibly licenced. Its the entire point of clean room reversing, to bypass copyright and licencing.

The training data is on the dirty room side, it doesnt get passed to the clean side.

The clean room concept is violated if raw source code is in the model generated, but to my knowledge its not. There is no way to pull out a chunk of the model, and say "This is a direct copy of main.c from xyz project". At best, you can ask the LLM to guess what it should be, and it'll regurgitate something statistically close. I just tried asking my local qwen3.8, and did produce a close summary of the nginx main file, but it wasn't even close to a 1 to 1 copy.

The clean room would also be violated if you consider the training, model and inference as a singular entity, but that would be tough to argue when the inference can happen on one server, the resulting model isn't functional in any way, its a big box of data, and the inference can happen on any GPU anywhere else in the world.

To me this is on the list of acceptable uses of AI: to fuck over asshole corporations.

A human doing a clean room reimplementation is not supposed to have seen the original source code for something they are reimplementing. If that source code is in the training data for the LLM, that seems like a similar situation.