Vibecoding Photoshop: Time and pressure
andreklein.net
andreklein.net
It was discussed here on HN yesterday and then, the anonymous person behind it removed the "Issues" section completely from their Github
https://news.ycombinator.com/item?id=49972730
https://replayweb.page/?source=https://archive.org/cors/gith...
If I may ask, if you were to do this all over again in 2024 - 2026, would you choose SSR?
Wouldn’t that screw up his whole thing? Because now you need infra to render to free users?
SSR only really makes sense when you either want to optimize for fast page loads, or do PHP-style conditional rendering or whatever
As another browser example along the lines of Photopea, ScreenRun is amazing as a true one-time pay, client-side app. I have loved that thing ever since I paid the $50. Such a cool app and model that has no server costs. Whoever made that is also my hero. But aren't those days now over? I guarantee there are >100 ripoffs of it now. Isn't an SPA now just basically open source, in the worst way only?
Meanwhile, Photopea.com only works via a website for normies. It is ad-supported, so given all the work put in, and updates, and the fact that it is web served... wouldn't SSR be a better option? (I hate this reality)
edit: I am allergic to Next, so I have been using Sveltekit since I realized that SPAs are now accidental open source. What have others been using for SSR and why? That is the discussion I would really like to have.
I guess my point was, how can anyone protect any client-side IP in the modern era?
And at 8 Euro a month I can't quite justify the Photopea premium plan.
I use it enough I would happily pay say $50 for a self-hosted single user license that will not get updates. But not 8 euro a month forever.
I am finally forcing myself to un-learn the decades of Adobe muscle memory and replace it with gimp/krita. Or maybe I'll vibe up something that does exactly what want and nothing more; bespoke one-off tooling is just so easy to do now.
I have a huge respect for your work on Photopea, been using it constantly since the time I completely moved to Linux a year ago. It has the best Photoshop parity yet to find! Would you ever consider to: 1. Open sourcing your work (partially or completely) under some kind of a permissive/copyleft license. 2. Rewriting some parts of the website that require performance as WASM targets. 3. Using any modern JS frameworks, e.g. Svelte.
1) About 30% of my original code is open-source as JS libraries: they end with ".js"-https://github.com/photopea?tab=repositories
2) WASM is not faster than Javascript "by default". Both can be fast or slow depending on how you write it. I don't think having everything in WASM would make Photopea significantly faster. Also, Photopea uses many libraries as WASM (WEBP decoding, ZSTD decoding, JPG decoding, Harfbuzz, ... JXL / AVIF encoding)
3) I never used any Framework, but if it is just a bunch of a JS code, I think I managed to write an equally good JS code for the UI over the past 14 years.
Can you settle an argument, perchance? Is it pronounced "Photo-P" or "Photo-pia" (as in "utopia")?
I had settled on the utopia sounding version but will try and think of peas now.
The immediate reaction is to ask whether any of these would be good enough to replace Photoshop (or Pixelmator or whatever). That's a first-order question and it's an open one --- you probably couldn't replace a real workflow with vibe Photoshop today, but I wouldn't bet against that being the case a year from now.
But the second-order questions include consideration of the people who can't use Photoshop, either because of cost or (to me, more importantly) because of lack of platform support.
In a world where 90% of the functionality of things like Photoshop and Word are replaceable with personal apps, how much does platform support matter anymore? How much do platforms matter anymore? I'm a committed macOS user and have been since the TiBook. Will I be 5 years from now? I'm deeply uncertain, but one thing I'm pretty sure of is that if I'm not a macOS user, I also won't be a Windows or Linux user.
Which is apparently impossible.
Who does Jony Ive work for now?
OpenAI Planning A 2028 Smartphone With Custom Chips
https://www.androidauthority.com/openai-smartphone-mediatek-...
Then there's things like properly sleep and suspend support.
The same applies to software: you'll describe a problem and a network of agents will write a solution faster than you can Google it, download it, install it, and start using it. It'll work with your accessibility requirements out of the box; it's UI will be in your language; the data will be stored where ever you like; and so on.
Instead of having a web application with a typical frontend and backend, I have fantasized about thin UI/UX interfaces with an almost pure LLM backend. Just having one or more bots receive parsed http requests, CRUD-ing the data based on training and tool access, and responding back if necessary.
(I have a strong dislike for web development, so perhaps my fantasies are a form of escapism.)
Please no. Enough with putting everyone into a box/label/profile. Have people be exposed to and interact with the multitudinal variety of the world!
If you think people are wasting their time scrolling reels now, the one social benefit is that you might have seen one of the same ones as your (human) friends. Imagine if that was gone. Truly horrifying.
This exists now.
You can have gemini AI - the that's built into google search - just make little applets. I just made one for a color picker, and generating palettes, including one that applies aces tone mapping. It showed made little adjsutable sliders and everything.
For a while, chatgpt web didnt count against usage, and you could use sol, it was isnane, you could upload a screenshot and just ask it to build out a little editor with threejs, and whatever else.
There's grok bot, dots, and other 'bots' that are just doing independent actions in the background. That can do digital actions like organize your documents, send emails for rebates, etc.
If you accept the shame, you can literally have bite sized story telling that are like mini soap operas, completely cheesy, predictable plots but that are very good guilty pleasures.
Now, you can just prompt the 10% you need and not even have to use up your computer's resources with the rest of Photoshop. I didn't like a single one of the many Android TV video player apps. Most of them were way too big, had all kinds of features I didn't need, took a while to launch, had stupid "libraries" that you had to populate your videos into, and relied on third party lookup services to get things like movie posters and other metadata. Pre-2020 I would have just picked the least bad one and lived with it. Now, I just vibe-slopped a player app with only the features I want and nothing else.
Out of curiosity, how much did that cost?
If an AI tool could possibly replace Photoshop with a prompt and a few bucks worth of tokens, do we really need the app at all - why not just ask the AI directly to do whatever it was you were going to do with the app?
But then one step further, why should any of us have jobs at all? If AIs can do almost anything a person at a keyboard can do, maybe 85% of the quality, but instantly for anything for a few bucks or less, is there any point in paying any of us at all to do anything we could do at a keyboard?
I wonder if it's going to come to that in 10 or 20 years.
I'll stick to gimp I guess.
But if I'm vibe-coding an app for myself, I only want it to what I want it to do. One product for one user.
I hate Photoshop's model as much as anyone but (pre-AI) solutions already exist which people already use.
I've said it in other venues, but we're reaching a point where using someone else's software is like chewing someone else's gum.
Or if you want to apply nondestructive edits to a photo, and you realize that PSD is actually a pretty safe forward-compatible option, you may want adjustment layers and a PSD writer.
I think most of the time people don’t want to use software, they just want their problem solved. But creativity is a notable exception.
If I have magic box that can give me anything, I'm not going to ask for tools and materials to build a house, I'm gonna ask for a house.
For now, while we are still riding this wonderful subsidized AI gravytrain, but how much will it cost in tokens/$$$ next year when investors finally want a return on their investment? Or if the AI bubble pops? I doubt running 7B models locally will suffice, even at the rate of improvement we are seeing.
For me, I just want to easily organize some folders, cull stuff, throw the culls into a marked directory, and process them to JPEG with some lighting adjustments since I shoot raw. Lightroom honestly is painful to use. I don't think gimp fills my needs since its more like photoshop. I'm happy to look at something else though!
Not yet production ready although working pretty well.
I based it off babl, GEGL and MyPaint. It is binary compatible with gimp through xcf and also supports svg natively.
Since I’m a Mac user it uses metal rendering for speedy painting, gradient and filter previews, then commits changes through GEGL asynchronously. It does work quite neatly.
It already found a couple of bugs in GEGL and Gimp, which I have tucked in patches and tickets. My plan is to go through these manually and do contribute the fixes, if the community accepts them (and if they’re indeed problems) However I don’t think an LLM should send patches.
Sure you can do a Photoshop clone as in "Ps in 1990" if you just ask an LLM because then (and for years after) Ps itself got all this stuff wrong and people used it, professionally nevertheless.
But getting these things right requires, as-always, in-depth understanding of the theory behind it or the LLMs will gladly omit those parts.
Not least because the majority of OSS code dealing with those things gets it subtly wrong. And they've been trained on that.
A good start is btw. to ask your agent of choice to copy all theory/basics from OpenImageIO.
How do I know? I'm maintaining a closed-source, heavily LLM-assisted Rust crate that is the backbone of all image processing we do at my employer (kringers fossed it will be OSS eventually).
The amount of subtly-wrong- bites-you-eventually bugs I fixed (or had an agent fix, rather), despite tight specs and good tests, over the last 10 months, is ... noteworthy.
That said, for a lot of users it probably doesn't matter if interpolating between red and green means getting a muddy brown or that you get subtely wrong colors when you downsample and some (or quite a bit ) of aliasing when you do so more than 2× etc.
But while a lot of people want to prompt a clone of app X and an unexpected many also seem to have the patience for it: few have the background to get sth. out of that endeavour that will be genuinely useful in the end.
The CAD/parametric 3D modeling segment is another category I'm watching closely where a lot of people have the same issue w/o (yet) realizing.
And ofc, there are always exceptions where seemingly someone is behind the project who actually do know what they're doing. :)
I can understand why you might write open source software for your fellow man but who wants to do it just for the benefit of OpenAI or Anthropic.
So Anthropic can vacuum it up and regurgitate it to you for $200 per month?
What does this mean for the American economy, 80% of GDP attributed to services? Will every country have their own ecosystem like China?
What stops Canada from decoupling from US tech, and making their own software ecosystem? Considering the Europeans were already moving in this direction, can't they just recompile their own versions of software?
Is the future of software just cloud software? Advancements in local encryption like TPM?
They will stream the software to you in video format, and you will send your inputs to the server, in this way the software code is not exposed to the user
You can still generate software via video, but novel implementations can still be protected this way.
More security restrictions yes, that will work, but unless you actually go airgap all the code, there's an access point somewhere which can be vulnerable, either through the network or through the human init?
I don't imagine that many of their customers are fans of taking unnecessary risks to save what isn't much money for most of them.
As a professional, there's no way I'd risk not using the real code. As a hobbyist, I can/have use something like Photopea.
If you're a casual user, just use AI directly - upload image and tell model to remove background, or make it cinematic, etc.
If you're a professional, you'll want the best.
I’ve decided to build my alternative because every time there is a prosumer tool I like (Pixelmator, affinity) they get bought out (affinity to canva), or the developers suck (Pixelmator developers cramming all features people wanted into a different “pro” version)
More info in another comment. One of the reasons I’m doing this is to also see how (and if) it is actually possible to build a complex software that I can use.
Tiny tools I’ve made tons, but something that needs actual architecting is different. Image editing is also a middle ground candidate. The output is what matters and is the final test. “Subtle” bugs will not ruin your day as long as you use a periodic backup
https://github.com/scosman/freecell
So far it’s GPU accelerated (GPUI), cross platform, supports 95% of excel functions, supports charts, is about 40x faster than Numbers, and can render 240fps.
I thought I’d leave this process with a new appreciation for manual dev/details. I haven’t found that yet. I’m getting more into architecture details, writing more benchmarks, testing more. I know there are limits, but I’m still trying to find them.
It would be absolute garbage if I wasn’t already and engineer. It probably is absolute garbage in ways I don’t know because I haven’t read the code. But the app is… quite good.
Then
>can this be done” at all? (...) The mere existence of these tools is proof that it can be done.
Then
>DISCLAIMER: I obviously can’t know whether any of these projects were definitely “vibe-coded”, but it also doesn’t matter
..????
Adobe Creative Suite Cleanroom Port to Rust
https://news.ycombinator.com/item?id=49981449
Photopea creator weighs in on Photosuite project
Opus et all are able to do everything needed client side
Which, taken in isolation, is BS. There are tons of good-for-nothing crapware that are a waste of time and space.
A functioning clone of the original software will be much better proof.
I’m forced to use Adobe’s apps because they are the golden standard at the moment but I wish with all my heart this to not be the case anymore.
Adobe fucking sucks.
Side-note to whomever attempts to build a new Photoshop: workshop better names. The ones on the article are woefully bad to the point of hilarity.
Why is their throne unearned? They created and continue to maintain their software, which is, as you said, still considered the gold standard.
And how are they in a monopoly position? Is this about proprietary formats or something else? Genuinely interested to know about this if it wasn't just hyperbole.
The throne was 100% earned. I would love to see Adobe dethroned now, but it definitely earned it in the first place.
"An open-source, clean-room reimplementation of Adobe Photoshop in pure Rust".
They also have: VectorCraft Vector illustration; FilmCraft Video editing, color and sound; LightCraft Photo library and raw development; PrintCraft Reading, organizing and protecting PDF; EffectCraft Motion graphics and visual effects; DesignCraft Page layout and publishing