Meta AI unveiled MusicGen, a groundbreaking AI model designed to create and mix music. This innovative tool is not only free but also readily accessible to users worldwide.
MusicGen, developed by Meta’s internal team Audiocraft, is an AI model that generates music based on text descriptions. The model can also utilize an external audio track to guide the creation process. It’s built on the Transformer model, a comprehension model created by Google.
The AI model has been trained on over 20,000 hours of licensed music, including 10,000 high-quality tracks, as well as stock music repertoires from Shutterstock and Pond5. This extensive training allows MusicGen to generate instrumental tracks up to 12 seconds long. Despite the current limitation in track length, the potential of this AI is immense.
MusicGen’s operation is simple yet fascinating. Users provide a text description, which can be supplemented by an audio file for melody extraction. The AI then generates a unique piece of music based on these inputs. It’s important to note that the AI performs the complete generation process each time it’s requested, leading to different results for the same description.
While the AI currently only creates 12-second song fragments, it’s expected that this limitation will be expanded in future updates. The model has been tested in three versions, with the most basic having 300 million parameters and the others having 1.5 billion and 3.3 billion parameters, respectively. Theoretically, the larger the model, the higher the quality of composition it offers. However, the intermediate level has been reported to provide the best results.
Techcrunch reports that MusicGen and Google’s MusicLM produce varying results depending on the input. MusicGen performs better in creating a Lofi electro song, while both generators show similar results in producing jazzy background music.
To use MusicGen, users can visit the Hugging Face project portal, input an English description, and click on the “Generate” button. An audio file can also be uploaded to influence the AI’s output. The generation process may take a few minutes, depending on the number of simultaneous users.
MusicGen is a testament to the advancements in AI and its potential to revolutionize various fields, including music. As AI continues to evolve, we can expect more innovative tools like MusicGen that push the boundaries of creativity.