Freepik, the online graphic design platform, has launched an “open” AI image generator called F Lite, trained on 80 million commercially licensed images. The model, developed in partnership with AI startup Fal.ai, contains around 10 billion parameters and was trained using 64 Nvidia H100 GPUs over two months.
F Lite is available in two versions: standard and texture. The standard version is more predictable and prompt-faithful, while the texture version is more chaotic and error-prone but delivers better textures and creative compositions. According to Javi Lopez, Freepik’s representative, the company has been working on F Lite for months, and it feels good to finally share it, as he stated on April 29, 2025.
The model was trained on an internal dataset of around 80 million images. The standard version is available at Fal.ai, and the texture version is also accessible. Freepik makes no claim that F Lite produces superior images to leading generators like Midjourney’s V7 or Black Forest Labs’ Flux family.
F Lite joins a growing collection of generative AI models trained on licensed data, including those developed by Adobe, Bria, Getty Images, Moonvalley, and Shutterstock. Running F Lite requires a GPU with at least 24GB of VRAM. The development comes as generative AI is at the center of copyright lawsuits against AI companies, with many creators and IP rights holders disagreeing with the practice of using copyrighted data for training without compensation.
Other companies are also developing media-generating models on licensed data, and the market could grow exponentially depending on the outcome of AI copyright lawsuits.
