In a candid conversation with VP Land, Joaquin Cuenca Abela, CEO of Freepik, lays out the story of a decade-long journey: from a humble image search engine to one of the largest AI-powered creative suites in the world.
Serving over 100 million users, Freepik has transitioned rapidly into generative AI, now producing millions of images and hundreds of thousands of videos every day. This article distills Joaquin's insights on that pivot, the product decisions behind it, the economics of AI generation, legal tradeoffs, and the company's vision for the near future.
From Search Engine to Creative Platform
Freepik began in 2010 as a focused solution for a common problem: designers and marketers needed free visual assets but searching the web was tedious. Joaquin, who had previously built a geolocated photography startup and worked at Google, returned to Spain and co-founded a service designed to find free images—essentially a Google for free assets. Over time that simple search engine expanded: Freepik started making its own images, added icons, illustrations, motion assets, and eventually built a full creative suite.
Growth was steady and, by Joaquin's description, comfortable—until 2022, when generative models like DALL·E dramatically shifted the landscape.
2022: The Moment the Future Shifted
Joaquin recalls his first encounters with modern image generation and how they forced a re-evaluation of the company's roadmap.
"Everything that we did in my mind became obsolete, or it was going to be obsolete in a couple of years. So to me, it was, at that point, AI crossed the threshold of having creativity."
That realization came in two stages. First, newer models produced images with composition and imagination that surprised even experienced creatives. Second, Joaquin faced the hard question: how could Freepik add value in a world where models can generate images on demand? The answer was pragmatic: iterate, integrate, and leverage Freepik's existing distribution and user base rather than attempting foundational research in-house.
Practical Integration: Start Small, Iterate Fast
Instead of building models from scratch, Freepik took an applied approach. They experimented with open-source models (e.g., early Stable Diffusion) and rolled out basic features: hosting models, adding curated visual styles, building faster generation flows, and tools like sketch-to-image. The key was focusing on user problems—making style selection simple, smoothing prompt crafting, and integrating generation into a familiar, productized UI rather than a raw model marketplace.
That product-first mindset became a competitive advantage. Freepik wasn't just another model host; it was a user-facing creative platform with a huge existing audience—over 80 million monthly active visitors at the time—allowing them to negotiate favorable inference pricing, funnel early feedback, and iterate quickly.
Scale, Economics, and Usage Patterns
Joaquin provided concrete metrics that help illustrate the scale and economics of Freepik's AI workloads:
Freepik serves over 100 million users across its properties.
The site has roughly 700,000 subscriptions on Freepik.com and more across other projects (over 800,000 total).
Daily generation rates are now about 3–4 million images per day and nearly 200,000 videos per day, placing the company among the top global generators.
Margins shifted as the cost of generation decreased rapidly. Image generation per-unit costs declined from a few cents to fractions of a cent, and Joaquin notes that gross margins for images rose dramatically as inference costs dropped. Video remains more expensive (roughly 20% margin currently), but improving model efficiency and usage patterns are driving better economics over time.
The Unlimited Experiment: Removing the Credit Friction
One notable product decision was removing the per-generation credit friction for higher-tier subscribers. Freepik introduced unlimited image generation for Premium Plus and Premium Pro members and ran rotating unlimited video weeks for those tiers. The rationale is both psychological and practical:
Charging per click makes users hesitate—akin to microtransactions inside a creative tool. Removing that friction frees exploration and experimentation.
By exposing unlimited generation initially to a small, paying cohort (roughly 10–15% of subscribers), Freepik could A/B test load and cost impact without risking the entire business.
Usage did increase (notably a 50% rise in total video creations when unlimited video access was toggled for that cohort), but it remained within predicted and managed bounds.
Joaquin emphasizes pragmatic guardrails: unlimited is "within reason," with device limits and anti-abuse measures to prevent single accounts from monopolizing capacity. The move also served as a conversion lever—some users upgraded simply to remove the psychological barrier of spending credits.
The UI Is the Product: Curating Models, Not Listing Them
Freepik treats the underlying models as infrastructure, not the product itself. Joaquin compares this to early personal computing: the CPU is hard to build, but the interface—the applications people actually interact with—become the product users care about. That philosophy drives Freepik's approach to model selection:
They do not indiscriminately add every available model. Each model must provide distinct, useful capabilities (e.g., better animation for illustrations).
The platform layers helper features—style presets, sketches, multi-input conversations—so users can achieve precise, professional outputs.
Eventually, choosing which model to use may become invisible—the system will orchestrate the best underlying model automatically.
Who Uses Freepik's AI—The Super Users
While consumer adoption exists, Freepik's strongest demand has come from professionals. Joaquin identifies three high-growth segments:
Marketing and design teams (the original Freepik core): designers working within marketing departments are the biggest group and still the backbone.
Visual effects and filmmakers: VFX artists in Hollywood and indie creators leverage AI for storyboarding, background enhancement, and production-level effects.
Product photographers and product placement: companies that formerly spent tens or hundreds of thousands on shoots increasingly use AI photography workflows.
Other growing pockets include architects, interior designers, and photographers—especially as tools like upscalers and enterprise workflows become more robust.
Enterprise, Legal Risk, and the Output Focus
Freepik offers enterprise contracts with indemnification (default $1M) because companies deploying AI at scale need legal assurances. Joaquin takes a pragmatic legal view:
European guidance has clarified that training on content found online is generally permitted unless a copyright holder explicitly opts out.
In the U.S., early rulings and interpretations are trending toward allowing content legally acquired for training to be used.
Freepik believes its partnered model providers are compliant; they actively monitor for instances where Freepik content (watermarked images, for example) appears in third-party model outputs and take action when necessary.
Importantly, Joaquin stresses that the most actionable legal concerns often fall on the output rather than the inputs. If a generated image clearly infringes—say, a recognizably copyrighted character used for promotion—then the output itself can be actionable regardless of how the model was trained. For that reason, Freepik implements alignment and moderation measures and works with enterprises to define acceptable commercial uses.
Where Freepik Sees the Next Wave
Looking ahead, Joaquin expects a split in applications:
End-to-end automated solutions will thrive where outcomes are measurable—e.g., marketing campaigns that can be A/B tested and iterated rapidly.
For artistic, one-shot creations—films, auteur work, and big-budget visual artifacts—creators will want precision, control, and a strong authorial voice. AI will be a tool that amplifies their vision rather than replaces it.
He also believes the product layer—version control, collaboration, shot tracking, storyboarding, secure enterprise storage—represents the next value opportunity. Freepik plans to invest heavily in these project management and workflow integrations so professionals can treat generative assets as first-class deliverables in complex productions.
Freepik and Hollywood
The company's tools are already finding their way into film and TV production pipelines, in use cases ranging from upscaling backgrounds to visual effects prototyping. Joaquin notes inbound interest from major studios and VFX houses; while some integrations are still in early stages due to legal and procurement processes, the trend is clear: AI tools are lowering the cost barrier for ambitious creators worldwide.
Final Takeaways
Freepik's pivot demonstrates a pragmatic playbook for legacy creative platforms facing disruptive AI technologies:
Recognize when a technology crosses a meaningful threshold (creativity, fidelity, or scale) and treat the moment as a mandate to experiment.
Leverage existing distribution and revenue streams to negotiate better costs and run controlled product experiments.
Prioritize the user-facing interface as the product—curate models and abstract complexity so professionals can focus on intent and outcome.
Be thoughtful about unlimited offerings: remove psychological friction for users but design anti-abuse protections and start with a controlled cohort.
Attend to legal risk pragmatically, focusing on outputs and enterprise indemnity while working with model providers to ensure compliance.
Joaquin's message is simple and sober: the future of creative work is different, and platforms that survive will be those that help creators keep control of their vision while lowering the technical and financial barriers to realization. For Freepik, that means continuing to refine the UI, curate best-in-class models, and build the enterprise-grade tooling that lets professionals treat AI-generated assets as mission-critical content.
