Matthieu Rouif was a product manager at GoPro in 2018 when the case for Photoroom turned up on his own screen. The marketing team needed visuals for an app feature, the designer was out of office, and he spent hours tracing edges by hand with a lasso tool. The conclusion he took from that afternoon, in his own words years later, was that the image is what makes or breaks a sale and everyone was treating it as an afterthought. The tools had not moved in a decade and they were built for designers, not for the people trying to sell something.

He was not new at this. Before GoPro he had built the first mobile apps for ski resorts, co-founded a social survey app called HeyCrowd, and led product on Replay, a video editor Apple named app of the year and GoPro later bought. What was new was the co-founder. Eliot Andres was an engineer who had been working on deep learning early, and the two of them left to build the thing Rouif could not find.

The first version went out fast. Writing about the tools he used, Rouif put a number on it: it took two weeks to build the first version of BG, from the first line of code on GitHub to release on the App Store. BG, later renamed Photoroom, did one job. It erased backgrounds, on the device GPU, quickly enough to run live on video on an iPhone XS neural engine. In May 2019 he posted a clip of exactly that, live background removal running side by side on three generations of iPhone.

Those two weeks are the anchor for the milestone on this page, and it is worth being exact about what they measure. Rouif has published the build window from first commit to store release, and he has published the stack that shipped inside it, including RevenueCat for in-app subscriptions. He has never published the date of the first paying subscriber. So the fourteen days here mark the moment the app was live in the store with subscriptions already wired, which is the earliest point a paying customer was possible, not a receipt with a timestamp on it.

The tools post is worth reading for what it says about the decisions he refused to spend time on. Fastlane and Figma to push the app and its screenshots into fourteen languages for a few hundred dollars. RevenueCat so that adding subscriptions did not double the development time, which is what he had watched it do at previous companies. Amplitude for the quantitative view of retention and funnels, Intercom for the qualitative one, and the same user id across both plus RevenueCat so paying users could be found in one and talked to in the other. Two engineers, two weeks, and an instrumented paid product.

Distribution came from the store rather than from a launch campaign. In November 2019 Rouif was thanking the App Store team for a spotlight on the PhotoRoom launch. By March 2021 he was presenting the app to Tim Cook and pointing out that a small team in France was serving customers in 175 countries with only three percent of revenue coming from France. Five months after that: growing fast, profitable, and more than a hundred thousand new users the previous day.

The obvious risk arrived on schedule. When Photoroom raised its Series A, Rouif says several investors were nervous that Apple had started building background removal into the operating system, and Meta then released Segment Anything as an open source segmentation model. Plenty of people concluded the problem was solved. His answer, published with a benchmark two years later, was about who the customer actually is. If you want to make memes, background removal is solved. If you have to process hundreds of thousands or millions of product images, five percent better quality and better speed still decides which tool you buy.

Photoroom raised a $43 million Series B in February 2024 with Balderton and Aglaé, alongside a foundation model built for commerce photography, and Rouif reported roughly three times growth over the previous year and the top spot for photo apps in more than fifty countries. By August 2025 the company was putting its own number on the result: past $50 million in revenue, and 300 million people who had used the product. In June 2026, announcing a rebrand under his own byline, Rouif wrote that Photoroom now processes more than 7 billion images a year for over a million businesses across more than 180 countries, and that the thing he is proudest of is not the scale but who it reached, from a two person sportswear brand to independent Etsy sellers.

The number he keeps returning to is not revenue but revenue per employee. He picked the metric up from Stripe, another Y Combinator company, and says that when Photoroom raised its Series A its YC partner told them he had never seen a company reach that level of revenue on so little capital and with so few people. That is the through line back to the two weeks. The first version was small enough that two people could ship it, priced from day one, and measured well enough to tell paying users from everyone else.