AI Is ‘Unstoppable’: Freepik’s CEO Discusses Artificial Intelligence, Rival Models, and the EU’s Limiting Approach to AI Regulation

Despite his busy schedule, Freepik CEO Joaquín Cuenca took the time to meet with Xataka and discuss his company, artificial intelligence, and his AI model’s main competitors for an hour and a half. We made the most of it.

Cuenca has strong opinions on all these topics, which comes as no surprise. He’s a prominent figure in technological entrepreneurship in Spain. After helping to launch classifieds site Loquo (acquired by eBay in 2005), he co-founded Panoramio, a geo-located photo mashup that was eventually sold to Google in 2007. After working with Google for three years, Cuenca co-founded Freepik with brothers Alejandro and Pablo Sánchez Blanes. 15 years later, Freepik has become a global leader in generative image AI.

Freepik’s AI Suite provides several options for creating and editing images and videos in various styles. It features models such as Flux, Google Image 3, and Veo 2.

Interestingly, Cuenca says he didn’t initially have an entrepreneurial spirit when he started university. “I was somewhat pulled into the world of entrepreneurship,” he explains. “I had always enjoyed creating things and had participated in open source projects and non-profit initiatives in the Linux community.”

However, it wasn’t the management side of companies that caught his attention, but rather the scientific and technological aspects. “I believe that my entrepreneurial spirit has emerged because the management layer in companies has been getting thinner,” he told Xataka.

Initially, he had considered a career in academia, but several factors led him to seek work in the industry instead. He began collaborating on different projects. One involved helping Eduardo Manchón, a high school friend who was working at Loquo at the time. eBay would eventually acquire Loquo, and Manchón encouraged Cuenca to work on something new.

It was during this period that Cuenca conceived the idea for Panoramio. He explains, “That idea began to grow, and I eventually quit my job to pursue it full-time. Google made us an initial offer, followed by a second one. We ended up selling the company, making it the first Spanish startup–small and somewhat anecdotal–that Google bought.”

This acquisition allowed Google to integrate Panoramio into Google Earth, where its usage skyrocketed. It also led to Cuenca and Manchón joining Google. Cuenca spent three years there and retains a “very positive experience.” However, it also clarified something for him. “At Google, there was always the understanding that if you do something, it has to be significant. For them, the minimum expectation was that you should aim for a billion dollars,” he points out.

Cuenca believes that “many projects in their early stages don’t seem capable of generating a billion dollars.” He cites Facebook as an example, “If you think about it, how many projects aimed at helping you connect with your friends have there been? There were plenty. What happened was that the execution was extraordinarily good.”

“It was the same with Twitter,” he adds. “It originated from a hackathon, but people questioned whether it would actually become popular.” For Cuenca, the message is clear: “Not every project that has an impact on the world appears to have this kind of potential at first.”

This perspective led him to think that Google’s strategy was prematurely stifling ideas. When asked if this was why Google shut down Google Reader, he replies, “That’s a good example, yes. The case of Google Reader is particularly evident.” He goes on to explain that, in a way, Google has demonstrated remarkable ambition and success with projects like Google Maps. The company built the maps “blindly,” using images and AI to gather all sorts of details, even without access to commercial maps for navigation.

However, according to Cuenca, Google has also made mistakes, particularly regarding cloud storage:

“We were one of the first major customers of Amazon S3. I chose it because I was running out of hard drives. One afternoon in Paris, I was transferring files to S3 while deleting others from my hard drives to make space for new incoming files. It was quite overwhelming. S3 saved us, to be fair. We were the only Google project using it. At the time, Google considered launching its cloud services but chose not to. They viewed their cloud technology as a key differentiator.”

Google chose not to offer it to the rest of the world to prevent others from gaining a theoretical advantage. However, Cuenca considers this decision a big mistake: “It easily set [Google] back two years” in the race for cloud platform dominance. This likely contributed to the fact that Amazon Web Services holds about 50% of the market share today, while Google’s share is around 10%.

Another instance where Google’s vision faltered and allowed a competitor to pull ahead was with Google Drive. According to Cuenca, “I think [Google CEO] Sundar Pichai was responsible. The week [Google] was set to launch Google Drive, he met with his engineers and convinced them to hold off on the release because he had an intuition that files and folders were about to become obsolete.” That prediction didn’t come true, and Dropbox managed to surpass it, although the market is now much more diversified.

More from Xataka On

Satya Nadella Is Skeptical About the ‘AI Revolution.’ The Reason Is Simple: No One Is Making Money Yet

Freepik and the Arrival of AI

When Cuenca first learned about AI, his initial thought was, “This makes what we’re doing obsolete. It doesn’t benefit me and makes my life more complicated. While it might enhance my capabilities, my first reaction is, ‘This changes a lot of what we can do.’”

However, Cuenca also realized something even more significant. “It was going to happen anyway,” he says. He dismissed potential legal issues that might slow down the evolution of AI models, believing they would ultimately be resolved through economic agreements. He also considered the ethical implications.

“I didn’t view it as unethical because, in general, what the machine produces isn’t a copy, even if there are extreme cases. People perceive different aspects, including general artistic styles, and it ultimately depends on the creator who uses that new tool to decide whether they want to create something similar to what has already been seen or something entirely new.”

This perspective influenced Freepik’s initial goals. “We started the company because we were creating websites, and it took us a long time to find the images we needed. We [initially] created it to find good images for free.” While there were cases where they couldn’t meet those needs, Cuenca explains, “AI is very well aligned with what we aim to achieve. In fact, it actually helps us expand the range of solutions we can offer our users.”

At every step, Cuenca was clear about one thing: “I saw it as unstoppable.” Although they debated whether to embrace AI fully or proceed more cautiously, they soon realized that they had to go all in.

This has significantly altered the operation of the creative sector that Freepik serves. “There are people who have been studying for years to acquire certain skills, only to realize that 90% of those skills have become irrelevant, leading many to seek help from psychologists,” Cuenca notes.

The new AI features have greatly impacted Freepik’s business. Cuenca provides a notable example. The company analyzes users’ first actions after subscribing to better understand why they did it in the first place. “We’ve reached a point where nearly 50% of new subscribers engage with AI as their first action. A year and three months ago, this figure was 0%,” he told Xataka. He adds that more than 50% of existing subscribers are now using AI on Freepik.

The evolution of AI image generators is impressive, and a similar advancement is occurring in video generators. Freepik recently announced that the company is the world’s first to integrate Veo 2, Google’s video generation AI model.

Cuenca encourages users to try out Freepik’s AI video generator, highlighting that there’s still much potential for development. “Video is currently where image generation was in 2023,” he points out. There’s still a lot of testing needed to achieve desired results, and “compromises must be made.” However, Cuenca says, “We’re getting closer to where image generators stand today: providing exactly what you want, understanding what you need.”

Midjourney, DALL-E and the AI Race

When discussing models competing with Freepik, it was inevitable to mention DALL-E, which Google seems to have halted. For this entrepreneur, the AI model “was a game for OpenAI.”

“If you have the opportunity in your company to get the best language model in the world, everything else is a distraction because it encapsulates all the intelligence. AI may be the most competitive sector in humanity’s history. It’s about trying to get intelligence. Whoever gets it will be number one and will have an advantage so huge that we can’t even understand it. With this type of potential, making images makes no sense at all.”

Cuenca had a different opinion about Midjourney. He stated that Freepik has “a lot of respect for Midjourney” but was clear that Freepik is superior in prompt compliance—how well it understands what you’ve told it. “Midjourney takes a lot of artistic license,” Cuenca said. According to him, Freepik is also superior for photorealistic images because “Midjourney has used a lot of images from movies, video games and a lot of anime.”

When asked how most image-generation models avoid certain types of uses—such as being unable to generate photorealistic images of famous people—Grok 2 and Grok 3, for example, don’t follow such restrictions and seem to have no censorship. Cuenca aligned more with xAI CEO Elon Musk: “I think we’ve gone too far.” For him, “telling an artist what they can and can’t do shocks me. It’s like giving someone a pencil and having them write or draw something you think is inappropriate.”

However, he explained that training-dependent biases can also cause models to generate inappropriate content. “We’ve done tests,” Cuenca said, “where if you train a model from scratch with images from the Internet and just ask it for the word ‘man,’ it will produce men in suits and ties, but if you ask it for ‘woman,’ 30% of the images it produces are naked women.”

In the end, AI is a tool. As Cuenca said, it can be used for harmful purposes, “just like the Internet, just like an encyclopedia, just like a chemistry class.”

Competition in AI and the European Union’s Choices

Cuenca explained that competition between models is fierce because these resources, he said, are “the fastest depreciating resources in the history of humanity. Something can go from being worth hundreds of millions of dollars to being worth nothing in six months, and it costs hundreds of millions to do it.”

For this reason, he believes the extreme competitiveness in this segment makes companies willing to lose money—a lot of it—because “in the end, whoever gets the highest volume will get the monopoly,” along with better margins and a cheaper model. He explained that a company with a lot of volume “will have to make its own GPUs and its own nuclear power plants” to lower costs and feed the vicious cycle.

Freepik is undoubtedly one of the reference companies in Europe for AI. For this reason, Cuenca’s views on what the EU is doing wrong—and whether recovering lost ground is possible—were particularly interesting. “There’s too much talk about the dangers, and the percentage of people doing ‘bad’ things is small, but the positive things that can be done are wild.”

He gave the example of Google and Meta, which are often criticized but, among other things, “have made it possible for every company to advertise. Before, there were very few channels, and they were only accessible to large companies. Thanks to Google and Meta, it’s possible to divide attention and advertise digitally to people who live in your neighborhood, which wasn’t possible before.” This example is striking because, while Google and Meta have allowed anyone to advertise, “Freepik can help these companies advertise in a more attractive way.”

According to Cuenca, “intelligence is the bottleneck for almost everything [invented in Europe]. The opportunity we have in front of us is incredible, and all they talk about [in Europe] are the problems.” In his view, the EU needs to be “more optimistic and see how much we can achieve. All the legislation that has been put in place avoids the exciting part.”

For him, focusing on AI’s dangers is a mistake. Almost everything has potential risks. “Fire is extremely dangerous,” he explained, “but we have to learn how to use it. Learning how to use it gives us much more than we lose from the fact that it is potentially dangerous.”

He said, “We need to change the mentality of the whole political class and make a 180-degree turn. [In Europe] they are beginning to realize that this change is necessary. The fact is that we have very good universities and some of the best researchers in the world, and they are all doing research and creating companies in the United States.”

He also believes efforts to create a Spanish or European AI are “nonsense.” He added, “It’s the most competitive sector in the history of humanity. If there are researchers who can do it [a private model], investment companies are more than willing to pay them. Why do public bodies have to do it? What business opportunity do they see?” he concluded.

Images | Joaquín Cuenca

Related | ‘On Google, You Get Links. On Perplexity, You Get Answers’: An Interview With the CEO of Perplexity, the AI Startup That Wants to Kill Its Biggest Inspiration