How Freepik Transformed From a Stock Image Platform Into a Generative AI Powerhouse

Fifteen years in the past, Freepik was simply one other inventory picture supplier, serving to designers discover the suitable visuals for his or her tasks. At present, it is a utterly completely different beast—a generative AI hub attracting over 60 million guests a month.

The shift wasn’t unintentional. It was the results of an organization prepared to rethink its goal and transfer past static photos to one thing a lot greater: a completely AI-powered artistic ecosystem.

Sitting in his front room , Freepik CEO Joaquín Cuenca Abela spoke with Decrypt concerning the state of the AI trade and the way his firm harnessed the chance generative AI introduced for digital artists.

“When generative AI appeared, we noticed we may develop our mission,” he informed Decrypt. “We have been now not restricted to serving to designers with pre-made content material. As an alternative, we may adapt to what they wanted and create one thing distinctive for every individual.”

The transfer paid off. Freepik is now a one-stop store for AI-powered picture and video technology, upscaling, animation, and extra. Cuenca Abela put it merely: “We simply need to give creatives extra management.”

From clean web page to AI engine

Freepik began with a easy premise: remove the frustration of gazing a clean web page. Earlier than AI, the platform supplied hundreds of thousands of inventory photos and templates that creatives may use as beginning factors.

“The slowest, most painful a part of the artistic course of was ranging from zero,” Cuenca Abela stated. “We helped remove that barrier by giving designers hundreds of thousands of photos they may start creating with.”

Now, with generative AI, Freepik doesn’t simply provide a library of content material—the agency creates it on demand.“A photographer wasn’t a conventional Freepik person,” stated Cuenca Abela. “They already had their photos. However now, with our upscaler Magnific, they will improve them in methods they by no means may earlier than.”

The shift has broadened its viewers past graphic designers. Photographers use it to reinforce, tweak, and upscale photos. Filmmakers experiment with AI-generated visuals, architects and inside designers construct ideas in ways in which have been as soon as time-consuming and costly, and the common Joe makes use of it to generate stunning waifus—as a result of, in fact, it’s AI we’re speaking about.

Not simply one other picture generator

Within the crowded AI area, Freepik is specializing in workflow integration. Most AI instruments specialise in one factor, be it picture technology, video creation, or upscaling. Freepik connects all of them, performing like a hub that integrates completely different open- and closed-source generative AI instruments in a single place.

Amongst different providers, the corporate’s AI suite consists of:

Picture technology with fashions like Mystic, Flux, Ideogram, and Google Imagen
Customized LoRA coaching for constant character and magnificence technology
Video technology utilizing seven completely different fashions, together with Google’s Veo2, Hunyuan, Luma, Kling, Hailuo, and Minimax
Modifying instruments for inpainting, outpainting, filters and seamless picture growth
Audio technology, together with music, voice-over, and sound results
SVG conversion capabilities for vector-based belongings.

One of many largest success tales has been Magnific, Freepik’s AI-powered upscaler. It went viral for its capability to reinforce picture particulars with out distorting them—one thing even top-tier AI fashions have struggled with.

Then, Mystic was the icing on the cake, offering outcomes that have been able to competing towards state-of-the-art fashions like Ideogram or MidJourney. Mystic is definitely a workflow, utilizing Flux as a core mannequin with quite a lot of tweaks behind the scenes.

That stated, quite a lot of consultants and lovers have tried to imitate Freepik’s secret sauce—which depends on open-source fashions—with combined outcomes. Freepik has a method to offer high quality outcomes, constantly, with the very best person interface doable, which is what prospects are paying for.

“Individuals typically underestimate the distinction between a superb product and a very good product,” Cuenca Abela stated. “The final 10% takes 90% of the hassle. That’s why many tried replicating Magnific, however could not fairly nail it.”

The copyright debate

With AI-generated content material comes controversy. Many artists argue that AI builders unfairly prepare their fashions utilizing copyrighted works with out permission. Cuenca Abela doesn’t dismiss their issues, however stated he sees the difficulty otherwise.

“When you required permission from each particular person creator to coach an AI mannequin, these fashions merely couldn’t exist,” he stated. “It’d be like asking permission to index each single internet web page earlier than launching Google.”

He acknowledges the strain.

“This within the brief time period damages the artist—utilizing one thing created by the artist. For the [affected] artists, this can be a scenario of profound injustice,” he informed Decrypt, recognizing that such development compelled them to evolve as enterprise. “One thing comparable occurred to us. When (AI) emerged, our enterprise all of a sudden introduced much less worth to the desk. We needed to adapt.”

That stated, he argues that AI-generated photos aren’t direct copies. “The grievance conventional artists normally have is that their photos have been used with out permission, which is completely true,” he admitted. “However the counterpoint is that the photographs these fashions produce aren’t copies. If an individual had made them, there wouldn’t be a declare of copyright infringement.”

The core of the controversy is mainly the trade-off between artistic management and technological progress. Cuenca Abela believes society will in the end favor AI’s advantages—simply because it did when comparable debates surrounded images killing portray, digital artwork killing conventional artwork, or web search engines like google killing encyclopedias.

“As a society, we’ll must steadiness issues and decide. If permission from the creator is required to coach a mannequin, generative fashions for textual content and pictures will not exist.” he stated. “[If that happens, then] society loses all of the progress that textual content fashions present. They might help us discover vaccines, medicines; the scientific developments they will carry are great. All that progress is misplaced.”

Cuenca additionally sees AI as a software for self expression. He doesn’t differentiate AI artists from artists.

“There’s no distinction. It’s a software. AI is a way to specific what you need, and artwork is the expression of what’s inside you, what occurred to you, your life experiences—effectively, you are able to do that with AI, with work, with photographs… It’ll rely on the artist,” he informed Decrypt.

“For me, it’s completely artwork and it’s professional. I don’t have an moral drawback with that.”

Open fashions vs. closed techniques

There was consensus about closed-source being the go-to choice for finish customers, with fashions and applied sciences normally being extra user-friendly and offering a better-quality expertise than open choices. Nevertheless, issues have drastically modified over time.

Secure Diffusion revolutionized AI artwork, Llama was key to bringing native textual content technology to the plenty, and extra lately, DeepSeek R1 reignited the controversy about closed-source AI corporations overcharging for his or her fashions.

Nevertheless, some customers nonetheless desire closed-source choices. Cuenca Abela has sturdy opinions concerning the AI trade’s future, significantly within the battle between open-source and proprietary fashions.

“When it comes to code, state-of-the-art open-source is on the similar stage as proprietary fashions,” he stated. “The largest distinction is coaching time and dataset curation, an extended post-training part, a bit higher tagging, and so on. However as for technical stage, I do not see a lot of a niche.”

Whereas proprietary fashions like MidJourney and Ideogram get extra refinement, Cuenca sees open-source alternate options closing the hole shortly. He factors to Flux for example: “It is perhaps a tiny step behind the very best closed fashions, however not two steps. And since it’s open, the neighborhood fine-tunes and builds on it, typically surpassing the closed variations.”

For Freepik, selection and adaptability are the priorities. “Somebody who is aware of the right way to use Freepik will get higher high quality than MidJourney,” Cuenca Abela stated. “When you want photorealism, we now have Google Imagen. When you want inventive textual content technology, use Ideogram. When you want character consistency, prepare a LoRA. No single mannequin is the reply to the whole lot.”

In different phrases, there’s no jack of all trades in AI. And the flexibility of selecting open and closed-source fashions on demand is extraordinarily essential to get the granularity required for the right murals—one that actually resembles what the person has in thoughts.

Freepik’s AI video wager

Not too long ago, Freepik has doubled down on AI-powered video instruments. The corporate built-in Google’s Veo 2, which dramatically improves video technology high quality.

“Earlier than Veo 2, you needed to generate 10 or 20 movies to get one which labored,” Cuenca Abela notes. “Now, with Veo 2, you get a superb outcome each different attempt.”

However the true game-changer for video artists might be an upcoming AI video editor, he stated. As an alternative of simply producing brief clips, customers will quickly be capable of assemble full movies fully within Freepik.

“At present you’ll be able to solely make video clips—solely generate small clips of two seconds, 3 seconds, 8 seconds. We’re engaged on one thing that permits individuals to edit them on the web page itself, add audio, and do the entire composition in order that you find yourself together with your clip,” Cuenca Abela informed Decrypt.

“The aim is to make Freepik the artistic hub the place you don’t want to go away the platform to complete a venture,” he stated.

The way forward for AI: Alternative or concern?

Are we near synthetic normal intelligence, or AGI? Will machines substitute us? Cuenca Abela sees AI’s fast growth as each thrilling and unsettling.

“[AGI] feels shut now—a lot nearer than anybody anticipated only a few years in the past,” he admitted. “We went from individuals dismissing AI as a toy to machines that may assume.”

There’s probably not consensus about what precisely constitutes AGI, however it may be broadly conceived as a sort of synthetic intelligence that may perceive, study, and apply data throughout mainly any discipline at a human-like stage or past, being able to adapting to new issues. We’re presently in a state of “slim AI,” with fashions that excel at some issues however underperform at others.

That shift, he argued, raises large existential questions. “Machines could be paused, restarted, or copied. People can’t. These variations matter,” he stated. “It means we’ll at all times have a novel place alongside expertise.”

Whereas some concern AI changing human creativity, Cuenca Abela stays optimistic about its potential. “I believe this can trigger a really profound and powerful acceleration that feels a bit overwhelming. We don’t know what we’ll be capable of obtain sooner or later.”

A bit extra right down to earth, he thinks the instant future might carry us extra instruments that assist machines perceive precisely what the person desires, being way more correct and offering higher-quality outcomes. And Freepik’s new philosophy seems to level towards that path, turning into a hub through which artists can discover the whole lot they should flip an AI technology into their very own imagined murals.

“That is our mission: Serving to individuals make nice designs to specific the ability of their concepts,” stated Cuenca Abela. “AI for us is only a software—however it’s how individuals work together with AI that issues.”

Edited by Andrew Hayward