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

Cyber Security, ICT, Most Popular, Trends News

No Comments

Photo of author

By Karla T Vasquez

WhatsApp Group Join Now
Telegram Group Join Now


Online Graphic Design Platform FreePic on Tuesday has unveiled a new “Open” AI image model that the company says that commercially licensed, was trained exclusively in “safe-safe” images.

Model, called F LiteThere are about 10 billion parameters – the parameters are internal components that produce models. F Light was developed with partnership with AI Startup Fall. And according to Freepic, 64 Nvidia H11 was trained using GPU for two months.

The F -Light Licensed Data Join the AI ​​models trained in a small and growing collection of AI models.

The generator is at the center of copyright cases against AI agencies including AI OpenAI and Mid Journey. It is often developed using a large amount of materials, including copyrighted materials from the public sources around the web. These models argue most companies that developed Fair use Protects their practice of using copyrighted data for training without compensation for the owners. Many creators and IP rights do not agree.

The Freepic F Light has created two flavors, standard and texture, both were trained on internal data sets of nearly 80 million images. Standard is more estimated and “instant-believer”, when the texture is more “error-prone” but according to the company, it provides better texture and creative composition.

An image of the standard model generated with the prompt is “a person standing in front of a sunset, in a glorious environment.

Fripic
An AI-exposed photo from the F Light Model of Freepic.Figure Credit:Fripic

Freepic does not claim that F. Light Midjorny’s V7, Black Forest Labs’ Fux Family or others like the top image generators create a higher image than generators. The goal was to make a model publicly available so that the developers could make it and improve it, according to the company.

It is said that running F Lights is not an easy fame. The model requires a GPU with at least 24 GB VRAM.

Other agencies who developed media-enforcement models in the licensed data include Adobe, Bria, Getty Image, Munvalei and Shutterstock. Depending on how the AI ​​copyright cases are shaking, the market may grow significantly.



Leave a Comment