With the rise in popularity of large language models (LLMs) and generative AI tools such as ChatGPT, developers have found use in a variety of ways to generate text, from writing emails to compiling article summaries. Now they want to help you generate music snippets just by typing a few words.
Brett Bauman, developer of PlayListAI (formerly LinupSupply), launched the new Songburst app on the App Store this week. The application does not require complex training. Just enter a query like “Soothing piano music to listen to while studying” or “Funky beats for a podcast intro” to have the app generate a music video.
Bauman revealed that he created the application’s backend using Vercel and the music is generated using Leap. There is currently a 30 second generation limit and some results may not be very good. According to Bauman, over time, he intends to increase the duration of the generated music video and improve its quality.
Songburst is free to try, but offers a subscription for $9.99 per month or $79.99 per year. The subscription gives 20 song credits per month and the ability to download tracks in mp3 format. Users can also purchase additional credits in packs of five ($7.99), 10 ($11.99), or 20 ($15.99) songs.
According to Bauman, he created this app because there are few simple and native mobile solutions for generating music from text that do not use spammy tactics to raise money for a subscription.
However, he is not alone in trying to create a neat text-to-music app. Ahil Tolani, the creator of apps like the music collaboration app Rapchat, has released CassetteAI, which is available both online and on the App Store.
At the input level, CassetteAI works similarly to other applications. You enter a request, and the application issues a track. However, it can generate a sample up to three minutes long. According to the creator of the app, this is because it runs on a custom model based on the seq2seq hierarchical architecture and is trained on a specialized dataset for generating non-copyrighted music.
The tool also provides an interface for users to create different versions of the generated tracks, edit them, and mix them to create a new track. These instruments are pretty basic, so don’t expect to create a multi-layered master track just yet.
Cassette AI also has a Pro subscription for $4.99 per month, which will give users access to unlimited compositions and access to better AI models to improve composition generation, Tolani said.
The developer noted that Cassette AI is better than other music generators such as Mubert and Beatbot because it generates better quality music and faster. He added that when creating Cassette AI, he wants to respect the ethical boundaries of the music industry.
“We want people to see AI as a tool for making music, not as a substitute for creators. Calculators have not replaced mathematicians, they have simply simplified calculations. We want to make music production accessible to everyone and for any use case,” he said.
These tools are mainly for creative people who can use copyright free music in their videos or podcasts. The developers also hope that musicians will notice their instruments and incorporate them into their sample or song creation process.
In addition to indie developers, large technology companies are also involved in the problem of turning text into music. Google introduced its MusicLM tool at the Google I/O developer conference in May. In June, Meta* made its own AI-based music generator called MusicGen publicly available.
Although the quality of the generated compositions is improving, there are concerns about the training data that is used to create the music. To avoid legal issues, OpenAI made its Jukebox model partially open and banned users from making music for commercial use. At the same time, there are AI-focused musicians like Grimes, who in April invited fans to create songs with her voice and share royalties with her.