Meta’s new AudioCraft AI can (allegedly) turn text into music

Mark Zuckerberg's Meta has just announced the launch of a new AI tool designed to create music based on text prompts. AudioCraft features three models — AudioGen, MusicGen, and EnCodec — which work for music, sound, compression, and generation. See below for an informative graphic explaining the kind of prompts they take.

Notably, all of the AudioCraft AI models have been made open-source, meaning that any users will have access to the source code. To avoid copyright violations resulting from data scraping, Meta had to train their AI using company-owned and specifically licensed music, but users will be able to train models with their own data sets of songs.