The last couple of years has seen a steady increase in the number of startups exploring AI music. The latest is Cassette, which is pitching itself as “your copilot for music”.

Co-founder Akhil Tolani has a long history in mobile music apps. He was the developer of iMusic, one of the most popular music apps in the early days of Apple’s App Store.

He’s since been the lead iOS engineer for music-creation app Rapchat, helping it scale up to 10 million users over the four years when he worked there.

So what is Cassette? It’s driven by Tolani’s research into AI music as well as co-founder Alex Meltzer’s experience growing companies in the web3 space.

“The main goal is to reduce the barrier of creating music,” says Tolani, expressing his ambition that Cassette can do for music what Midjourney has done for art – while stressing that replacing professional musicians is not on his agenda.

Cassette is being deliberately careful not to box itself into any one particular usage or customer group.

“It’s a great way to make some beats to study to, or for your home video, or just creating a mixtape for someone,” says Tolani.

“We’ve had people wanting to use the platform for amateur film production or video-game compositions. A father wanting to make a soundtrack for his wife. People wanting to teach kids about music…” adds Meltzer.

Until recently, the standard way to interact with this kind of AI-music tool was fairly limited. You’d choose a length, a style/genre and maybe some parameters around that, then click a button and see what came out.

Cassette, like the recent wave of generative AI services, is driven by text prompts, with some of the most successful shown off on its homepage.

“Craft a mesmerising and captivating hip-hop beat that seamlessly flows with an introspective vibe. Start with a slow hypnotic piano melody, delicately played with melancholic chords, evoking a sense of reflection,” is one example.

“Begin with mellow drums, gentle kicks, subtle snares, and relaxed hi-hats. Layer in soft piano chords or soulful guitar loops, deep bassline, and subtle textures like vinyl crackles. Keep melodies simple, harmonious, and evocative. Maintain a relaxed tempo.”

Tolani says that using text prompts means the experience of using Cassette can be “more boundless” than previous tools of this type. There are some limitations though: telling the system to create music like a specific artist won’t work.

“We tell people that the model will not understand who Drake is, or who Daft Punk is, because it is not trained on their music,” he says.

“It would understand ‘give me a hip-hop beat for an emotional tone’ or ‘give me reggaeton music mixed with dubstep’. But if you ask it to mix Drake with The Weeknd, basically it doesn’t know who Drake or The Weeknd are.”

That’s clearly important for another reason: the music industry’s current unhappiness with deepfaked tracks based on, yes, exactly those two artists has led AI music startups falling over themselves to declare that they don’t train their models on copyrighted music.

“Every single record label will sue you if there is even a slight overlap in one of the output samples!” says Tolani. “They are waiting to sue you. ‘That’s been a boring weekend, who do we sue on Monday?’ We have to be very careful in that situation: we cannot output something that even coincidentally matches the output made by an artist.”

So what does Cassette train its AI on? Sounds.

“We have trained the model on short bursts of sound samples: audio clips between 20 and 30 seconds, that are more towards the core instruments, genres and concepts in music rather than full songs,” says Tolani.

He adds that Cassette is looking to partner with a third-party service that will be able to compare its AI’s output with the commercial music catalogue “so that we can guarantee 100% that the music output will be copyright-free”.

Like other startups in its field, Cassette is walking a tightrope between its ambitions to democratise music-making, and its desire not to be seen as something that negatively impacts professional musicians.

“Saying ‘our AI will be able to make the next Beatles song’ is not what we’re aiming for over here. We are not aiming to replace musicians,” says Tolani. “The calculator did not replace mathematicians. It just made it easier to calculate things!”

“I want to build a music production platform where the consumers don’t even care that it is powered by AI. I want it to be a platform for someone in college who has a presentation tomorrow and wants to make some background music for it, or for people in an early-stage rapping career who wants some beats to experiment with.”

As it prepares for launch, Cassette is also raising an $800k pre-seed funding round, of which around a third has already been committed at the time Music Ally spoke to Tolani and Meltzer earlier this month. Cassette is also in talks with labels about potential partnerships.

Alongside launching and fundraising, Tolani wants to publish some research papers outlining the technology behind Cassette, particularly its ability to generate longer pieces of music.

“Being able to generate a three-minute beat in five minutes? That’s realistic,” he says. “No one’s going to wait 45 minutes just to generate something longer than 45 seconds.”

EarPods and phone

Tools: platforms to help you reach new audiences

Tools :: Wyng

Through Music Ally’s internal marketing campaign tracking, we’ve recently discovered an interesting website by the…

Read all Tools >>

Music Ally's Head of Insight