Freepik, the well-known online design platform, has launched a new AI image generation model called F Lite, which stands out for being trained entirely on licensed, safe-for-work content. Built in partnership with AI startup Fal.ai, F Lite boasts approximately 10 billion parameters and was trained over two months using 64 Nvidia H100 GPUs. The company emphasizes its commitment to legality and ethical standards, especially as the generative AI industry faces intense scrutiny over the use of copyrighted data without permission.
Most generative AI models currently available, including those from companies like OpenAI and Midjourney, rely on datasets scraped from the web — much of which may include copyrighted material. These practices have sparked legal challenges and ongoing debate, with AI developers citing fair use as justification, while many creators and copyright holders disagree. In contrast, Freepik has built F Lite using an internal dataset of around 80 million licensed images, positioning it as a more transparent and compliant alternative.
F Lite is offered in two versions: standard and texture. The standard version is designed to be more prompt-faithful and stable, while the texture version, though less consistent, offers richer textures and creative outputs. Freepik clarified that its goal isn’t to outperform top-tier models like Midjourney V7 or Black Forest Labs’ Flux, but rather to provide an open model that developers can modify, adapt, and use freely for further development.
Despite its accessibility in terms of licensing, the F Lite model demands significant computing resources, requiring a GPU with at least 24GB of VRAM to run. Freepik’s approach aligns it with other companies like Adobe, Shutterstock, Bria, Getty Images, and Moonvalley, which are also working to create AI models trained on rights-cleared content.
