Freepik has officially launched F Lite, a new Freepik AI image model developed in collaboration with AI startup Fal.ai. Announced Tuesday, F Lite is positioned as a transparent and customizable generative AI tool, trained solely on commercially licensed, safe-for-work images. The model enters a competitive landscape amid growing concern over copyright and data use in AI training.
Built with Ethics in Mind: Licensed Training at Scale
F Lite was trained using a dataset of approximately 80 million images, all internally sourced and commercially cleared. With 10 billion parameters, the model was built over a two-month period using 64 Nvidia H100 GPUs, according to Freepik.
The company offers two distinct versions of the model:
- Standard – More accurate and prompt-faithful.
- Texture – Allows for richer visuals but with a slightly higher margin of error.
This bifurcated model structure gives developers the flexibility to choose between predictability and creativity, depending on their use case.
A Response to Copyright Controversies
F Lite enters the market at a time when generative AI companies like OpenAI and Midjourney are under legal scrutiny for training their models on publicly scraped — and often copyrighted — content. In contrast, Freepik aims to avoid such issues entirely by building its model from the ground up using only licensed media.
This makes Freepik one of a growing list of companies — including Adobe, Shutterstock, and Getty Images — that are prioritizing ethical training data in generative AI development.
Developer-Friendly, But Demands High-End Hardware
Although Freepik is not positioning F Lite as a direct competitor to high-end tools like Midjourney V7 or Black Forest Labs’ Flux, it is openly available for developers to fine-tune, explore, and innovate on.
However, there’s a catch — the model requires a GPU with at least 24GB of VRAM to run effectively, which could be a limiting factor for casual users or smaller teams.
Get the Latest AI News on AI Content Minds Blog