In the rapidly evolving landscape of artificial intelligence, where innovation is constant, a significant development comes from Freepik. Known for its extensive library of graphic resources, Freepik has just unveiled something that could reshape how we think about AI image creation and copyright. For those tracking the intersection of technology and digital assets, this new AI model offers a fascinating case study. What is the New Freepik AI Model? Freepik has launched an “open” AI image generator called F Lite. This model stands out because, according to the company, it was trained exclusively on commercially licensed, safe-for-work images. Developed in partnership with AI startup Fal.ai, F Lite contains approximately 10 billion parameters, the building blocks of the model’s intelligence. The development process involved training using 64 Nvidia H100 GPUs over two months. This focus on licensed data positions F Lite within a small but growing group of generative AI models that are specifically addressing the complex issues surrounding data sourcing and intellectual property. Why is Training on Licensed Data Important? The training data used for generative AI models is a major point of contention in the current legal landscape. Companies like OpenAI and Midjourney face copyright lawsuits because their models were often trained on vast amounts of data scraped from the public web, which includes copyrighted content. While many AI developers argue this falls under fair use, creators and rights holders frequently disagree, demanding compensation or permission. By training F Lite exclusively on their internal dataset of around 80 million commercially licensed images, Freepik aims to navigate these legal challenges proactively. This approach offers a potential path forward for developing AI tools that respect existing copyrights and provide clearer provenance for the training data. Exploring the F Lite AI Model Variants Freepik has made two versions of the F Lite model available: Standard Version: This variant is designed to be more predictable and faithful to the user’s prompt. It aims for accuracy in interpreting instructions. Texture Version: This version is described as more experimental, potentially more error-prone but capable of generating better textures and more creative compositions. It allows for artistic exploration beyond strict prompt adherence. Both versions were trained on the same core licensed dataset, offering users different approaches to AI image generation based on their creative needs. How Does F Lite Compare to Other AI Image Generator Tools? Freepik is clear that F Lite is not necessarily intended to outperform leading models like Midjourney V7 or Black Forest Labs’ Flux family in terms of raw image quality or artistic style out of the box. The primary goal for releasing F Lite openly is to provide a foundation for developers to build upon, tailor, and improve. However, running F Lite is not lightweight. It requires a GPU with at least 24GB of VRAM, which means it’s not accessible to everyone. This technical requirement highlights the ongoing challenge of making powerful AI models widely available. The Growing Trend of Generative AI on Licensed Data Freepik is not alone in pursuing the licensed data approach. Several other companies are also developing media-generating models using carefully sourced and licensed content. This group includes notable players like Adobe, Bria, Getty Images, Moonvalley, and Shutterstock. The increasing focus on licensed data suggests a potential shift in the AI development landscape, driven by legal pressures and the demand for ethically sourced AI tools. The future growth of this market could be significantly influenced by how ongoing AI copyright lawsuits are resolved. A legal framework that clarifies the use of copyrighted data for AI training could either accelerate or hinder the development of models like F Lite. Conclusion: A Step Towards Responsible AI Image Generator Development? Freepik’s release of the F Lite AI model trained on licensed data is a noteworthy event in the generative AI space. While it may not instantly surpass established players in artistic output, its foundation on commercially licensed content directly addresses critical copyright concerns facing the industry. By making the model “open” (in the sense that developers can potentially work with it), Freepik aims to foster innovation while adhering to a more responsible data sourcing model. This move highlights the industry’s ongoing efforts to balance rapid technological advancement with legal and ethical considerations, potentially setting a precedent for future AI image generator development. To learn more about the latest generative AI trends, explore our article on key developments shaping AI features.
