Javi Lopez and Emilio Nicolas were not first time founders and they were not in a hurry. The two had spent thirteen years building Erasmusu, a student housing site out of Murcia in southeast Spain, and had sold it. By Lopez's own account, both of them had then been living what he calls a sabbatical life for two years when generative AI arrived. He marks his starting gun as April 2022, when DALL-E 2 came out. He had no financial pressure and all the time in the world, so he spent roughly eighteen months reading Reddit threads, following open source image model developers, running tests, and posting what he learned.

That posting habit matters more than it looks. He opened his X account a year before Magnific existed and used it purely to share generative AI tutorials and experiments. He ran a Substack called Latent Space with the same material, going back to February 2023. In his own retelling, he had already built the acquisition channel without knowing that was what it was.

The product came out of the experiments. Somewhere in the tests a workflow emerged that did not just enlarge an image but redrew it, inventing detail that had never been in the original file. Midjourney output at the time was low resolution and creators were desperate for a way to make it printable and usable. Lopez had been quietly running his own posted artwork through the workflow for three months before anyone knew what the small script in his images was.

Development took about three months and, by his account, under $5,000 to get started. Neither founder needed a salary because the previous sale had left them a cushion. About half of Magnific's source code was written with Cursor and Claude, which in late 2023 was unusual enough that Lopez says people did not believe him.

They launched by throttling. For two weeks before opening the doors, Magnific ran a waitlist and let only a handful of creators in. Those creators posted before and after comparisons, and the comparisons did the work. Lopez describes people offering the early users money per image just to jump the queue: "can I pay you for each upscale you make for me?" That is about as clean a demand signal as a founder gets before charging anyone.

Magnific opened to everyone on 27 November 2023, announced on the product's own newsletter. There was no free trial, and the reason was blunt: the workflow is GPU heavy and they could not afford to give it away. Anyone who wanted to use it bought tokens. Lopez later said they had margin from day one. The first customer milestone here is anchored to that opening day, 27 November 2023, because the product was paid from the first hour it was public.

A week later they relaunched on Product Hunt, after a first attempt broke on Product Hunt's side, and ran a 24 hour testing day giving everyone 50 free tokens so people who had been priced out could see it work.

What followed was not a campaign. The founders never bought an ad. Five months after launch, in May 2024, Lopez published the numbers: 725,000 registered users, a two person team, zero investment, zero paid campaigns. In the same post he announced that Freepik, the Spanish stock asset company, had acquired Magnific. He had told whoever asked that he would sell only to someone who shared their values, and both founders stayed on after the deal.

The revenue picture came out later and in pieces. In February 2025 Lopez wrote on LinkedIn that Magnific had gone from $0 to $10M ARR in one year with only two people. In April 2026, when Freepik rebranded itself as Magnific, he put figures on it: $10.2M of revenue in 2024 and $34.42M in 2025, the second number combining both sites after the acquisition, at a 90% gross margin after GPU costs.

That leaves one gap we cannot close honestly. Lopez has never published the date Magnific crossed $100,000 in annual recurring revenue, and given a paid only launch that went viral it was almost certainly weeks rather than months. The only anchor his own words provide is the outer bound: one year after launch, ARR was $10M. So the $100k ARR milestone below is dated 27 November 2024, exactly one year from the launch date, and should be read as the latest it could have happened rather than the earliest.

The replicable parts are duller than the outcome. Two experienced operators with savings spent eighteen months getting good at a technology before building anything with it, published in the open the entire time, made the product paid on day one so demand and revenue were the same signal, and refused to hire. Lopez is direct that the timing was luck. In his words they were an ant on an inflating balloon, and he does not think he could repeat it.