Eight days. That’s how long it took for seven free Adobe look-alikes to pass 42,000 GitHub stars between them, according to a write-up by Pasquale Pillitteri that counted the numbers on October 8. The apps are PhotoCraft, VectorCraft, FilmCraft, LightCraft, PdfCraft, EffectCraft and DesignCraft, and they imitate Photoshop, Illustrator, Premiere, Lightroom, Acrobat, After Effects and InDesign. One developer named Brandon Thomas wrote them in Rust with Claude Opus 5.5, and HotHardware says he started because he’d paid one cancellation fee too many.
I tried to read the Reddit thread where people were fighting about this. It says “removed by moderator” and nothing else, so I’m going on the ArtCraft site, the GitHub readmes and the news coverage. I’m telling you upfront because the story is already getting stretched in both directions. Some people say Adobe is finished. Others say it’s a toy that crashes. Both camps skip the more interesting part, which is that copying a product like this used to need years and a big team, and that cost was doing most of the protecting without anyone noticing.
Reddit Link: Discussion removed by Mods, But has Links
Website Link: Seven apps. One craft.
The law didn’t move, but the cost of copying did.
What got built, and what didn’t
Details first, because they matter. PhotoCraft opens and saves layered PSD files, and the project claims 134 of 135 real test files round-trip byte for byte, which Gadget Pilipinas reported. The interface reuses Adobe’s menus, panels and shortcuts on purpose, so if you know Photoshop you already know where things are. IT-Connect says eight Claude agents running in parallel produced about 70 menu items in a few hours. The apps are free under MIT and Apache 2.0 licenses, they run on Windows, macOS and Linux, and some of them also run in a browser tab through WebAssembly.
Now the other side. Mezha reported that Thomas first promised full parity with Adobe in a month, then changed it to 99% in under a year, and that the real number today is more like 25 to 35 percent, with users complaining about basic editing, shortcuts and speed. PhotoCraft had 447 open issues on October 8. One of them is a crash when opening a 32-layer PSD. A Photoshop replacement that dies on 32 layers isn’t a replacement yet, since plenty of working files are bigger than that. The project labels PhotoCraft and PdfCraft “early alpha” on its own site, so to be fair it never pretended otherwise.
One more thing about how it was built bugs me. Pillitteri counts 24 crates and over 1,700 tests in PhotoCraft, and 35 crates and more than 1,600 tests in FilmCraft, where a crate is just a module of a Rust project. IT-Connect adds that the AI ran many of those tests itself to check its own work. Big numbers, sure, but a test written by the same model that wrote the code mostly checks that the model agrees with itself. I spent way too long trying to decide if that’s a real weakness or just me being old-fashioned about testing, and I still don’t know. The 447 open issues suggest the tests missed plenty, because a crash on a 32-layer file is the kind of thing a real user finds in five minutes.
Also, I keep saying “clone” and that’s a bit unfair. The team calls these clean-room reimplementations, and that word matters a lot once lawyers show up. Actually, “reimplementation” oversells it too at 25 percent, so call it what it is: a very fast first draft of seven products.
Why the law didn’t slow this down
Clean-room design is an old trick. In the early 1980s, when Compaq wanted a PC that could run IBM’s software, one team studied the IBM BIOS and wrote a description of what it did, and a second team that had never seen IBM’s code built something to match the description. It was slow and it needed lawyers in the room, and the cost was the point, because it kept the practice rare. The funny part is that it built a whole industry of IBM-compatible PCs, and nobody calls that theft today. Whether ArtCraft ends up remembered like that or like a court fight is still open.
The ArtCraft version looks different. PhotoCraft’s readme says it was built from public specs and observed behaviour only, and Adobe really did publish the PSD file spec, which is why Pixelmator and Photopea can already open those files. But one developer plus a model trained on a huge pile of public code and documentation is not two separate teams.Pillitteri’s write-up adds that nobody independent has verified the clean-room claim. I can’t tell you if a court would accept it. I’m not a lawyer, so treat all of this as a layman’s reading.
The bigger issue is what copyright covers at all. In 2021 the US Supreme Court ruled in Google v. Oracle that Google’s copying of Java API declarations was fair use. Copyright protects the text of code, and what a program does mostly sits outside it. If your product is basically a set of behaviours, a model that has seen those behaviours can write fresh code for them. PetaPixel made a related point, as summarised by Ecosistema Startup: this legal setup grew up when reverse engineering was so expensive that undocumented code worked like a lock. The lock was the price tag.
And there’s a twist I only noticed late. The US Copyright Office has said that material produced purely by AI, without human authorship, can’t be registered, which may mean the clones’ own code is harder to protect than their licenses suggest. If so, somebody can copy the copy.
The real legal exposure for ArtCraft, as Gadget Pilipinas points out, is trade dress, because matching an interface this closely could get challenged. As far as I could find, Adobe hasn’t sued. If it does, that ruling will decide how close anyone is allowed to get to a famous interface, and every future clone will be living inside the answer.
It isn’t only Adobe
Look at what happened to software stocks in early February 2026. After Anthropic released Claude Cowork plugins, investors panicked about AI reproducing popular tools, and a Jefferies trader named Jeffrey Favuzza called it the “SaaSpocalypse” when he spoke to Bloomberg. A Reuters headline, quoted in a Rothschild & Co newsletter, said nearly $1 trillion had been wiped off software and services shares. The iShares software ETF had its worst stretch since 2008, down more than 14 percent, according to the Motley Fool. Around the same time DeepLearning.AI’s newsletter noted that Lee Robinson at Cursor had written in December about replacing Sanity, a content management system they paid for, with a custom setup built from scratch. That’s a paying customer deciding the product is something it can just build.
The selloff half-reversed, though, and that tells you something. DeepLearning.AI reported that shares bounced a bit the following week when Anthropic partnered with some of the same software companies, but they didn’t get back what they’d lost. So even the market couldn’t decide if AI was the copier or the partner. I find that more honest than either headline. It means the people with money on the line don’t know where “copied” ends and “built on” begins, and that’s a legal gray zone as much as a financial one.
Games got hit first. In the last week of January, Google showed Project Genie, which makes interactive environments from text prompts, and shares of game developers sold off sharply, per the same Rothschild newsletter. Nobody was surprised, really. Games have always been the easiest thing to copy. When Flappy Bird blew up, app stores filled with lookalikes within days, and the mechanic was so simple that nobody could own it.
The sharper example is Palworld, and it doesn’t involve AI at all. Nintendo and The Pokemon Company sued the studio Pocketpair in Tokyo in September 2024, over patents on creature-capture mechanics. They weren’t going after character designs, which surprised me when I first read it. In November 2025 they narrowed the case to older versions of the game that Pocketpair had already changed, and Nintendo Everything reports that if Nintendo wins it would collect only 5 million yen, roughly $30,000. A related Nintendo patent was also rejected in the US, according to coverage listed by OpenClassActions. Palworld 1.0 shipped on July 10, 2026 with no legal block. A court hearing for evidence was set for October 1, and I couldn’t find any report on what happened there, so as of today that part is still hanging.
Think about that for a second. A company with real patents, real lawyers and two years still can’t stop a game that borrowed its core loop, and Pocketpair built its game the slow human way. Now remove the labour cost from the copying side.
There’s an odd twist in the Palworld story as well. Mein-MMO argues the lawsuit gave Pocketpair momentum, because every new court headline kept the game in people’s feeds for two years, and GameVexo says player sentiment leaned heavily toward the small studio. So the big company’s legal threat worked like free advertising for the copy. I don’t think that’s an accident of one case. Players mostly don’t care who owns a mechanic, they care if the game is fun and if the company suing looks like a bully. An IP claim that makes its owner look bad is a weaker weapon than it used to be.
Everyone has a fair point
The developer’s view is simple. Thomas has said everything can now be open source, cross-platform and written in Rust, and Mezha reports that he says ArtCraft’s existing business can pay for the free apps, so there’s no need for a subscription. Adobe’s side is messier, because the anger isn’t only about IP. HotHardware notes that Adobe recently settled a $150 million federal lawsuit over subscriptions that were hard to cancel, so a lot of the energy behind these clones is people being fed up with the billing, not people wanting to fight over copyright. That matters, since you can’t fix a trust problem with a patent.
The working designer’s view is different again: Ecosistema Startup says freelancers and studios who depend on PSD, AI and InDesign files can’t just move, because their plugins, their habits and their clients’ files are all built around Adobe. Code is one layer, but the files and workflows people are stuck in are another layer, and a free app doesn’t touch those on day one. The investor’s view has real numbers behind it too. A Gartner report from February, quoted by Orion Advisor Solutions, expects $1.4 trillion of software spending in 2026, which is 15 percent more than 2025, so companies may buy more software, not less.
And then there’s the engineer’s view, from Calcalist: a flood of AI-written code in 2026 will need serious cleanup in 2027. I tend to believe that one, because I’ve never met a codebase that stayed clean after a fast sprint.
The “it’s only alpha” argument is true today and weak as a plan, since alpha is a stage, not a place you live. Starting from zero, ArtCraft got to roughly a quarter or a third of Photoshop’s features in under two weeks. Nobody should assume the next quarter takes two weeks, but nobody should assume it takes ten years either.
A small tangent, because it’s been in my head. A lot of designers in India learned Photoshop on pirated copies, and almost everyone knows it. My guess is those users later became paying studio customers, whether Adobe planned that or not. Free clones with proper licenses feel like the same story with paperwork, and I’m not sure which side it helps.
The small maker gets hurt most
One thing the Adobe story hides is who loses the most. Adobe has lawyers and a brand everyone knows, and it can sit through a two-year case. Nintendo can too. A solo developer who builds a good little app can’t. IP always worked best for people who could afford to enforce it, and now that copying is cheap, the maker with no legal budget is the easiest target.
Picture a one-person app that took eight months to build and has a small, loyal user base. Somebody points an agent at the public screenshots and the help pages, and a free lookalike appears the same week. The original has no registered trademark and no money for a lawyer. That version of the story worries me more than Adobe’s, and I haven’t seen many people write about it. Maybe I’m just not reading the right places.
The counterpoint is that small makers also gain, because they can now copy the giants, and ArtCraft is the proof. True. But copying up and being copied aren’t the same trade. Copying a giant gets you attention and GitHub stars, while being copied as a small maker can end your income, so I don’t think the gains and losses are equal.
What’s left to protect
IP on features, layouts and “how it works” is close to dead as a moat in practice, even if it’s alive on paper. If you ship something people love, assume a version of it exists on GitHub within a month, written by somebody with an agent and a grudge.
What survives is a shorter list. File formats that people are already locked into, which is why Adobe publishing the PSD spec was a bigger decision than it looked. Data that nobody else has, which Orion’s note calls hard for AI newcomers to copy. Trust and support, since a studio isn’t going to bet a client deadline on an app with 447 open issues. Brand names and trademarks, where ArtCraft was careful: the apps carry no Adobe names, and a NOTICE file keeps ArtCraft’s own name and logos outside the free license. And speed of shipping, though that one’s the weakest because a model can ship fast too.
If I were at Adobe, I wouldn’t sue first. A lawsuit against a free open-source project whose Photoshop version has 23,303stars would make the cancellation-fee story louder, and that’s the story people are already angry about. I’d fix the cancellation flow and keep the file formats open enough that they stay the standard. Then I’d ship AI features the free apps can’t match, though that last part is a guess, since I don’t know what Adobe has in its vaults and it could be wrong. The order is what matters here: trust first, lawyers last.
Some people will say customers always pay for quality. That works only while the quality gap stays wide, and the gap just got a lot narrower than anyone planned for.
Which is also why the 32-layer crash matters more than the star count. If you want to test all this yourself, download PhotoCraft this week and open the biggest PSD you have. It will only take a minute. Then try again in a month and see if the file opens.