About this site
Pixmunk is built by one person. It exists because moving AI models into the browser makes a genuinely free, genuinely private image tool possible for the first time, and almost nobody had done it. This page covers who is behind it, how it pays for itself, and what it will never do.
The short version
Pixmunk is five image tools that run AI models inside your browser instead of on a server. That one architectural choice is why the tools are free with no limits, and why your photo is never uploaded. Those are not two separate features. They are the same fact stated twice.
Why it exists
I kept running into the same annoyance. You want to cut the background out of one photo. You find a tool, upload your image, and discover the free version gives you a 0.25 megapixel preview and wants $0.20 for the real thing. Or it wants your email. Or it stamps a watermark across the result.
None of that is greed. It is arithmetic. Those tools run their models on GPUs they rent by the second, and a free user costs them real money on every single image. Once you see that, every credit system and sign-up wall in the category makes perfect sense. They have no choice.
Then browsers got WebGPU, and the tooling for running models client-side got good enough that real ones could run at usable speed. That changes the arithmetic completely rather than incrementally. If the model runs on the visitor's machine, the marginal cost of their ten-thousandth image is identical to their first: nothing. Unlimited and free stops being a loss-leader with a trapdoor in it and becomes an accurate description of the cost structure.
The privacy angle came along for the ride, and turned out to matter more than I expected. When the model runs locally, "we never see your photo" stops being a promise you have to trust and becomes a property of the architecture. There is no upload endpoint on this site. I could not receive your image if I wanted to. That is a much stronger thing to be able to say than any privacy policy, and it is verifiable in about thirty seconds.
Who builds it
I am Michael Machatschek. I am a senior AI engineer, based in Brussels, and I have spent my career building machine learning systems in production. I also run a small studio of consumer web apps, which is where I learned most of what I know about what people actually want from a tool versus what is fun to build.
This site is a side project, built and maintained by me alone. That has obvious consequences worth being upfront about: there is no support team, no SLA, and no 24-hour response time. What there is, is someone who will actually read your email and who understands the code end to end.
If you want to check any of that: my LinkedIn is here. It is a fair thing to want, given this site is asking you to believe a strong technical claim about what happens to your photos.
How it makes money
Right now it does not. There are no ads, no analytics, and no accounts.
The plan is ads, placed below the result rather than around it, and lazy-loaded so they cannot wreck the page's loading performance. That is a modest business and it is the honest one for this design: because inference costs nothing, the only real bills are hosting and bandwidth for the model files, and a plain ad-supported site covers those comfortably.
There may eventually be paid extras that on-device makes cheap to offer, like batch processing or a desktop wrapper. The five tools on this site will stay free and unlimited. Not as a promotion, and not as a trial. There is nothing to meter, so there will be no meter.
What will never happen, stated plainly so you can hold me to it: your photos will never be uploaded, sold, or used for training. That is not a values statement, it is a description of a codebase with no upload route in it.
What this site will not do
Some deliberate omissions, since a list of what a product refuses to do is usually more informative than its feature list.
It will not use non-commercially-licensed models. Several of the best models in these categories forbid commercial use, and plenty of "free" tools ship them anyway. RMBG-1.4 and MODNet are the obvious examples in background removal: both are good, both are off the table here, and no amount of quality changes that. A site built on a license violation has a deadline on it.
It will not pretend to beat the paid tools at everything. remove.bg has a better API. Cleanup.pictures' Pro model handles large removals better than our object remover does. The comparison pages say so, in a table, with the rows where they win marked as wins. A comparison page that claims a clean sweep is an advertisement, and everyone can tell.
It will not add an account system. There is nothing to store.
It will not hide the slow path. If your browser has no WebGPU, the tools run on WebAssembly and are genuinely slow, and the interface says so before you wait rather than after. Pretending otherwise would just make people think it is broken.
Get in touch
If you find a bug, a wrong claim, or a licensing mistake anywhere on the site, I want to hear about it, and the last one especially. Email is the best way: michael@1000ventures.io.
If you find a way to make the no-upload claim false, that is the most valuable email you could possibly send me, and I will fix it the same day.