FreePik, online graphics design platform, unveiled a new “open” “Open” image form on Tuesday, which the company says it has been trained exclusively on commercially licensed images and “Safe-For-Work”.
The form, called And LightIt contains about 10 billion parameters – parameters are the inner components that make up the model. FTe was developed in partnership with AI FAL.AI starting and training using 64 NVIDIA H100 graphics processing units for two months, according to Fribak.
F -Lite joins a small and growing collection of the Trucitomic IQ trained on licensed data.
We have secretly worked on this for several months! It is good to share it finally!
Links:
• Ordinary version: more predictable and guided, but less technical: https://t.co/mywsker9ir
• Texture version: It is more chaotic and at risk of error, but it provides better strength and … pic.twitter.com/gx5mipye8o
Javi Lopez ⛩ (Javilopen) April 29, 2025
Tructured artificial intelligence at the Center for Copyrights against Artificial Intelligence Companies, including Openai and Midjourney. It is frequently developed using huge amounts of content – including copyright content – from general sources about the web. Most of the companies that develop these forms argue Adel use It protects their practice of using copyright data for training without compensation for owners. Many creators and intellectual property rights holders do not agree.
FreePik has provided flavors of FITE available, standards and textures, both of whom were trained in an internal data collection of about 80 million pictures. The standard is more predictable and “fast”, while the texture is more “vulnerable to error” but provides better strength and creative installations, according to the company.
Below is a picture of the standard model that was created by demanding “a person standing in front of the sunset, in a majestic perimeter.”

Freepik does not claim that F Lite produces superior images on leading photo generators such as Midjourney’s V7 or Black Forest Flux Flux Family or others. The goal was to make a model available publicly so that the developers could design and improve it, according to the company.
However, the operation of the Fite is not easy. The model requires graphics processing unit with at least 24 GB of VRAM.
Other companies that develop media generation models include licensed data Adobeand Wildand Gety picturesand MoonvalleyAnd Shutterstock. Depending on how publishing lawsuits occurred in Amnesty International, the market can grow dramatically.
https://techcrunch.com/wp-content/uploads/2022/08/ai-gen-unfiltered.jpg?resize=1200,675
Source link