Freepik releases an ‘open’ AI image generator trained on licensed data

Freepik, the online graphics design platform, presented a new AI image model on Tuesday, which, according to the company, was only trained on commercially licensed, safe images.

Called the model F liteContains around 10 billion parameters – parameters are the internal components that make up the model. F Lite was developed in collaboration with AI Startup Fal.ai and, according to FreePik, was trained with 64 NVIDIA H100 GPUs in the course of two months.

F lite joins a small and growing collection of generative AI models that are trained on licensed data.

We have been working secretly on it for months! It feels good to finally share it!

Left:

• Regular version: more predictable and faster, but less artistic: https://t.co/Mywsker9ir

• Texture version: is more chaotic and more susceptible to errors, but delivers better textures and … pic.twitter.com/gx5mipye8o

– Report lopez ⛩️ (@javvilopen) April 29, 2025

Generative AI is at the center of copyright lawsuits against AI companies, including Openai And Midjourney. It is often developed from public sources on the Internet with massive quantities of content – including copyright -protected content. Most companies that develop these models argue fair use Switches your practice to use copyright -protected data for training without compensating the owners. Many creators and IP rights holders do not agree.

Freepik has made two flavors from F Lite available, standard and texture, both of which were trained on an internal data set of around 80 million pictures. Standard is more predictable and “promptly loyalty”, while the texture according to the company, information from the company, provides better textures and creative compositions, but provides better textures and creative compositions.

Here is a picture of the standard model that is generated with the command prompt “a person before a sunset in majestic environment”.

Freepik does not claim that f lite creates pictures that are superior to the leading image generators such as Midjourney’s V7, Black Forest Labs’ Flux family or others. The goal was to make a model openly available so that developers could adapt and improve it according to the company.

Apart from that, it is not an easy task to run f lite. The model needs a GPU with at least 24 GB VRAM.

Other companies that develop media -generating models for licensed data AdobePresent BriaPresent Getty picturesPresent MoonvalleyAnd shutter stick. Depending on how Ki -Urhleberrechtslagen, the market could grow exponentially.