Tech giant Meta is developing four new AI models to assist creators and developers in curating new apps. The models are available for public use and the design has been published in a document delineating the latest models and their potential future use. The rise in AI demand has led to the production of AI apps using these models, further enhancing their capabilities. Meta’s latest effort includes two new Chameleon variants, as well as JASCO and AudioSeal models.
JASCO is designed to accept various audio inputs and enhance sound quality. Users can adjust features like bass sounds, acoustic elements, and melodic tunes to create a great beat. The model can also accept text inputs for flavoring music.
Examples of the models in action involve creating tunes related to Blues, jazz, or rock music and ensuring that the listener is aware that the music was generated using AI-based apps through the presence of watermarks. AudioSeal goes a step further by indicating which segments of the tune were generated through AI, with watermarks accurately placed in those locations.
Meanwhile, the two Chameleon models can convert text to visuals and may be rolled out with limited features. Both variants, 7B and 34B, require models to understand pictures and technology, enabling them to carry out reverse processing and generate captions for images.
Source link
Source link: https://www.digitalinformationworld.com/2024/07/metas-research-team-is-working-on-four.html