A free, open-source 12MB Photoshop alternative just dropped. This is amazing and unsettling. Computer use has rendered industrialised app cloning possible with relative ease, this is the new web scraping. I suspect apps like Photoshop will protect against computer use soon much like we have DRM for games and movies, although it's probably already too late. This is what happens when paradigm-changing technology evolves so fast it evades all friction with the law. Most users only need the most common 5-10% of Photoshop's features. However, before we get too excited, Photoshop is more than 1,000 applications built into one. It's bloated and slow as hell, but (miraculously) it works. All of the complexity lives on the long tail. Extrapolating that it's possible to implement the full version of current Photoshop using AI frankly misunderstands the complexity of Photoshop. It's not that it would be impossible to implement (or partially so) with AI, but it would still take around 5 dedicated expert years and cost around 20 million dollars in frontier token costs alone assuming the prompter had deep enough conceptual knowledge of Photoshop. It's not a given that you would even make that money back assuming it was implementable in a robust/performant way with today's economics. Same deal with UnrealEngine, Davinci Resolve, Blender and other advanced professional software, only they are all even more sophisticated than Photoshop. However, there has never been more uplift to build tooling, or proxy tooling for specific purposes, and this can often be done by a single technical person in a short span of time and a handful of Pro subscriptions. Also, missed opportunity to implement agentic/learning Photoshop with a CLI interface 😆
Robbie Tilton
@robbietilton
Sep 18
I built a Photoshop-like image editor. It’s called Compositor. robbietilton.com/compositor I originally built it for myself to get off my Adobe subscription, but decided to release it for free and make it open source. It has all the essential tools I need for compositing, with none of the BS. I know this workflow is kind of archaic in 2026, but I’m still using it until AI can actually get pixel-perfect on some of the details. The entire app is 12MB. Photoshop is 6,455MB on my machine.

Last edited Sep 19, 2026 · 8:57 AM UTC

12
5
58
8,559
Sort replies: Relevant Recent Liked
Replying to @MLStreetTalk
I cloned Photoshop last month without any computer use with just telling it what I wanted and some screenshots. In the process I figured out new features that fit my workflow. Anyone can now create their own software to fit there needs. Try it!
1
21
I am. I've made a video editor! I probably wasn't very clear in my post. I meant the computer use provides a rich training set that allows the model to understand how to create something like Photoshop when folks like you prompt it to do so. If you look at one of my other replies, I explained it in more detail.
1
42
more replies
Replying to @MLStreetTalk
"Computer use has rendered industrialised app cloning possible with relative ease" Can you explain why computer use is the key factor? Seems like the simpler Photoshop use cases this app covers would have already been solved many times.
1
54
Absolutely! The basic thesis is that in order to clone something like Photoshop, the models need to acquire an abstract understanding of how it works. As you know the models start by pre-training just on text on the internet and then we do a bunch of post-training. The reason why post-training gives the model a more abstract understanding is because it is interactive. Similarly computer use is mixing visual data and action trajectories that give the model a much higher-level causal understanding of how applications work. Every time anyone uses computer use on their systems, agentically automating applications or building applications, that data is being fed back into training and helping OpenAI and Anthropic develop a deeper causal understanding of many popular applications. The hard part of software engineering is drawing the big arrows: the system design, the user interface, the domain model, the high-level guiding constraints that form the joints or the scaffolding of an application. The net effect is exactly what you see with the original post. Now you can ask it to create something akin to Photoshop and the model will, without much detailed prompting, just understand at least one previously established and known working version of what to do. All you need to do is give the model a screenshot of Photoshop and it will invoke an entire nexus of deep understanding of Photoshop built from all of that training data.
1
51
Replying to @MLStreetTalk
Yet the incumbents have a ton of path dependencies and rebuilding the project from ground up in many ways is easier, once you know the destination. Photoshop is an evolved piece of software mwell over 3 decades and just as with biological evolution it's hard to change course.
1
89
This was the point I made at the end of the post. It's easy if you take one known path. Photoshop is not one path, it's one thousand paths that diverged at different points in time. I agree with the spirit of what you're saying, though, which is that it's often easier to start again from scratch. That's only possible when you have acquired a new abstract mental model that integrates paths together, which in many complex cases simply isn't possible.
2
87
Replying to @MLStreetTalk
You know Blender us free already right?
1
59
Yes, I'm saying it's very sophisticated/ highly evolved
71
Replying to @MLStreetTalk
Blocking computer use would only accelerate their demise. At this point it would not take years or 5 experts. A single or two person team of dedicated engineers/experts could do it in less than six months and I think with the models coming soon, maybe only a few months. It's not just vibe engineering and it's not just one shot, but a sufficiently advanced coder/designer can do this now. And how many of those features do people truly need? I know the old joke that "everyone uses only 5% of Word, but everyone use a different 5%." That's true. But if you really think about it companies used to sell one off versions of software because upgrades were *compelling* enough because they brought legit new breakthroughs and features and workflows. That failed. And they forced the subscription model on us to stay alive but core features have atrophied in most apps in recent years. I still have an Adobe subscription and I use the new version and I cannot tell you the difference in the app in any of the last few years other than generative fill that is awful. To me the core innovations of the pen tool, masking, layers. magic wand, select shape, brushes were invented a decade ago and haven't changed.
184
Replying to @MLStreetTalk
@mlstreettalk 这事儿真是让我心情复杂啊！你说它惊艳吧，确实，一个12MB的东西就能替代Adobe全家桶里的大块头，这简直像从未来穿越回来的黑科技。可细想又觉得后背发凉，这克隆应用的速度快得让人害怕，以后那些大公司还怎么活？感觉就像街边小吃摊突然被全自动机器人抢了生意。 说到这儿，我突然想起以前大家
6
Replying to @MLStreetTalk
I remember this was considered Sci-Fi just back in April
Thomas Ahle
@thomasahle
Apr 15
Jensen says tool makers, like Power Point, will be more valuable when AI agents 1000x their user base. But what prevents the agents from just copying those tools? We may see more tool makers create an agent moat around their tools. Your agents won't be able to use the tools directly, but have to talk to the PowerPoint agents. This is similar to many high paid professions today, like Lawyers or Doctors. They don't give you direct access to their tools, and that makes it harder to study and copy them.
1
79
Replying to @MLStreetTalk
12mb photoshop clone via computer use as the new scraping is unsettling in a good way
64