Most of the new Stable Audio 3.0 offerings are free to download and build upon, allowing users to customize their own models.
Stability AI has announced the launch of Stable Audio 3 — a family of three new AI music models and one audio-based special effects model. Most of these new releases are “open weight,” a term used to describe models that are publicly available and can be downloaded and built on for further development, and are created via licensed training data.
Related
According to the research paper released alongside these four models, they are trained on a combination of licensed audio from production library AudioSparx (806,284 audios) and Creative Commons recordings from Freesound. To ensure no authorized copyrighted material ended up in the dataset from the Creative Commons portion of the data, music recordings were filtered using a tagging tool.
Key improvements from Stable Audio 2.0 to the new releases include variable length generations, allowing for up to six minutes of song to be created, and full song composition on portable devices like phones and tablets.
