Hacker News Reader

Comment by rf15 | Hacker News Reader

Tested it: very broken software produced by AI. The title and description is so far away from the truth you may well call it deceptive.

View on HN

Very late to this but for what it's worth I stumbled upon this and tested their release from earlier today and it seems basically fine. Everything I've tried works except for popping out panels (like the history panel, layers panel, etc) which doesn't appear to be supported. Not sure what was broken for you since you neglected to elaborate, but if things weren't working for you right up front then it must be iterating quickly. Not saying this to brag or to take a stance on it or anything, just observing.

Yes, I build and ran the photoshop clone on my Mac just now, and it seemed mostly fine.

For the problems I did face, I am not sure if they were bugs in the software or just me not being familiar with Photoshop with the Mac's touchpad controls.

I'd think this project kind of settles whether one can build large, working software with AI, for those that were still unconvinced.

I am surprised this has not seen more discussion on HN. The post is flagged.

Every piece of beta software that has ever existed, has bugs. AI doesn't change this. Software development has largely changed to build with AI first, get basic things working, iterate and fix. This is just the reality older developers just have to accept.

Bugs are ever present. And beta software should be treated with charity. But Photoshop, let alone the entire suite, is a beast. An effort, like this says it is, aiming for parity fundamentally misunderstands the reality of both the software they’re trying to replicate and the capability of AI assistance now and in the foreseeable suitable future

Indeed — what I am reading of this strongly suggests a misunderstanding about exactly what it is that makes Photoshop what it is.

I think it might be possible to slop-code an Excel clone because — truly minor interactivity quibbles aside — it can be tested with a truly exhaustive suite. The clone can be proved to do what it is supposed to do, over whatever subset one would choose to implement.

But Photoshop is not an application in the same way that Excel is an application.

(Not least because while powerful, Excel, like Word, is an implementation of an existing idea. Photoshop is the OG; it has all the history of the OG)

Photoshop is also a surface and a medium. It has a feel, it has responsiveness. It's like oil painting or watercolour: its own thing, with its own pen/brush dynamics, its own interactivity.

Part of the reason people don't want to switch away to Affinity (even though Affinity is free as in beer, highly capable and in some ways better) is just that to a set of very hardcore users, it does not have this feel. They just don't like how it feels.

Automated testing that focusses on pixel-precise rendering misses the point. It may actually matter less that it renders PSDs pixel-accurately than that it actually feels right.

If you are not convinced by my argument, look at "dodge and burn retouching". For better or worse, high end fashion photo retouching is done in an extremely precise, wildly technical, tiny subset of Photoshop tools.

If you want to replace Photoshop it has to be responsive at that level. And since that level is partly subjective experience, it's going to take some people.

oh c'mon, this thing IS garbage but it's completely possible to build a good one with AI, it just takes craftsmanship which is the part the dude spent no time on.

Not with the kind of vibe coded automatic testing they are talking about, though.

You really need to sit down with a Photoshop user to see how they feel about it and how they approach alternatives; it took me no time to adapt from Photoshop + Wacom tablet to Affinity Photo + Wacom tablet, but I know photographers who won't because it's "not right".

Serif had more luck converting Illustrator users to Affinity Designer, and made the right decision to release Designer first, because Illustrator users were much more frustrated with Adobe, and because giving those users what they wanted helped get past the way the app does not "feel" exactly the same.

There is no way that a "pixel-accurate compliance" test suite can capture what it is that makes these applications the way they are.

to make a complete copy of photoshop you just need computer use, which this clearly didn't do.

did you test it? the photo application opened a photo, pinch to zoom worked, the menus are full of buttons. it seems like the features i tested worked. they are mostly in similar spots as photoshop/photopea.

its a little impressive that models have a compressed approximation of photoshop inside that works with minimal tinkering.

the icon for windows doesnt work.

I tested it, I did different things and nothing worked right.

Office apps are strictly based on written spec. You can if spend good amount of token could easily build something very close not but 100% with that spec. PS is a different animal.

I'm struggling with the audience? The main thing about Adobe right now is millions of hours of youtube videos that show me how to do stuff. I would be completely lost with this software

I haven't had a chance to use it, how are you lost?

If you have an LLM and MCP why do you need a tutorial? Why would I watch a tutorial when the LLM can tell me how to use a feature? There is not a moat just because there are tutorials. That moat is gone. I don't need to watch tutorials anymore.

You can pass a youtube video of a tutorial directly to LLM, LLM can then use MCP to do what the tutorial says.

The more interesting part is that people "release" this "products" and market it even if were like version 1.

honestly, for what it is, it aint too bad, to think ai did this in this short of a time, im actually stoked on what the project could become.

I want to leave adobe for good and this LightCraft thing does what no other alternative can, Presets from lightroom.

if the creator fixes Masks and uses local ai possibilities for masking and such, this could very fast result in me canceling my sub.

So far yes its a vibe-coded mess but your crazy telling me adobes hella broken software is better.

Be sure to provide feedback for the project, actual field experience is the most valuable thing you can have right now.

Really? The creator said this three days ago

> Software is over. But for real. I just one-shotted Photoshop. The whole thing.

You're telling me it's not the entirety of photoshop?

surprised Pikachu face

I wasn't ready for this to be posted on HN, but this is moving so fast.

99% parity will take a while, but I'm sure it'll be measured in months and not years.

Feel free to check in on this claim. It's measurable.

you posted this on multiple subreddits and claimed "We're going to reach 100% parity within a month."

and now it's not ready for HN? what happened in the last 24 hours for that to be the case?

i look forward to a beta release of all 7 softwares with 99% feature parity november 5th.

Reddit isn't HN.

I needed to seed a community that could help contribute bug reports and PRs.

I had no intention of posting a thread here until it was more mature, because I knew this would be the response.

For what it's worth, kudos. This is very good use of ai, and bun proves that it is weeks to months till this hits stability.

The haters aren't seeing the direction, and proven examples.

In what way does Bun prove this?

Bun was a conversion of the source code to a different language.

what proven examples? lmao

The direction is going to be mostly about money and legality, I'd have thought.

This will certainly be a good test of the latter. Adobe are very wealthy, and presumably one of the enterprise customers AI companies want to get.

Vibe coders just drain them money and will never pay API prices.

But who will win?

> I needed to seed a community that could help contribute bug reports and PRs.

Holy shit dude you cannot possibly just say out loud that you're using people.

I know that AI makes you feel like god, but this is the style of communication that will be quoted against you. If you truly want to play this game (which I _highly_ recommend you shouldn't) then you must absolutely step that up.

why? Microsoft used Vista to test Windows 7.

seems completely normal to release free beta software and ask for feedback.

Disregard the haters. They were here yelling about how bun rewrite would never work and yet it's done in a couple of weeks.

Thank you.

There are some incredibly smart people on HN, but I wasn't quite ready for this project to face the firing squad. It's still much too new for the world's harshest critics.

With lots of additional effort, I expect this to succeed much like bun. Fingers crossed.

I sympathize.

Obviously you don't need to pay for Photoshop anymore. Does that mean your product is the replacement? No. Just an example.

The fact that you or anyone can get to 95-99% very quickly. The rest can be added over time at a very high pace and some "actual effort".

Unless Photoshop can add some sort of proprietary features then why use them? (they have been doing this with AI features mostly, but it should still all be copyable)

> The fact that you or anyone can get to 95-99% very quickly.

fact? what fact? have people lost their fucking minds here?

This is not even 1% of Photoshop yet. They won't get to 95% in a lifetime or 10.

the "photoshop equivalent" is really broken at a fundamental level. I can't do ANYTHING I tried to do. I'm sure it sort of works but the basics aren't there. Have you tried using it yourself?

The Bun rewrite is qualitatively different, is it not? It's a source to source conversion.

Yes, it is, but it was deemed impossible as well.

by all the same people who are questioning this slop or ?

Haters and people that are denying the reality, so basically flat earthers at this point.

Did the Bun rewrite work? I'm seeing some absolutely terrifying bugs in the GitHub issues, although for all I know they were in the old code, too.

Looks like. Claude Code uses it.

I mean Claude Code is one of the most buggy piece of software in existence, so I wouldn't use that as a measurement. There's also massive marketing forces in play here that can't be ignored.

You posted this in a comment on HN three days ago: https://news.ycombinator.com/item?id=49934844

And then four days ago:https://news.ycombinator.com/item?id=49922904

And you've mentioned ArtCraft numerous times https://hn.algolia.com/?q=by%3Aechelon+artcraft

Obviously you've been trying to get traction on this for a while. You can't claim that you "one-shotted photoshop" and then drop links and act surprised when someone decides to test that claim.

I didn't post a thread, ffs. I don't know why you're all jumping me, haha.

Calm down and be patient. Let the community token spend have time to diffuse into the products.

This effort needs a community to climb the quality and parity gradients. But I'm not ready to really show it off until we've hit some thresholds.

I've got the attention I needed to bootstrap this, so I can go back to work.

Bro, you don't get to ask for calm and patience when you make claims like "software is over. But for real. I just one-shotted Photoshop. The whole thing." and then post your repo in various subreddits and HN comments. You did ask for this, whether you have the self-awareness to realize it or not.

But look on the bright side, this is a chance to prove all of the haters wrong. Given your enthusiasm about AI and its capabilities it shouldn't be an issue.

I see, I was wondering why you did not post it to HN and only mentioned it in comments.

Makes perfect sense to me, since this is obviously an early alpha version.

It already looks very impressive to me, and I look forward to the project proving these doubters wrong.

I hope there will not be legal trouble with Adobe.

Would be interesting if you honestly documented as to if/when it starts breaking old things as you add new features.

scbench.ai suggests this will happen on projects where the human is no longer in the loop, and it's a good sized project to find out.

Reply on news.ycombinator.com