Whenever I look at a new generative AI model that can make music based on text prompts, I test it immediately to see what it can do. Most of the time, the results are incredible, but the samples are too short, and each one requires manual playback.
I thought that would be the case with Riffusion’s new Fuzz AI model. I hit play and went to make coffee while the music sample was playing before actually looking around to see what the deal with Fuzz was. But something else happened.
Instead of a basic UI, Riffusion offers a music player experience with lots of AI songs to choose from. They’re full-length tracks that you can stream just like you would with Apple Music and Spotify. The difference is that Riffusion doesn’t have ads, and you don’t need to pay a subscription to listen to or generate songs.
The actual playback stopped me in my tracks. My mind was blown and I nearly forgot about making coffee. I picked a random category, and the music was amazing. I realized right then and there that if someone told me this was an upcoming song from a new artist, I would have believed it.
Best of all, it’s totally free, as Riffusion said on X that the music generation tool will be available to users free of charge “for as long as our GPUs survive.”
“Introducing FUZZ – a generative music model like no other,” Riffusion said on X on Thursday. “Personalized, full-length, high-quality, and infinite. We’re making this instrument free for as long as our GPUs survive.”
The thread offers various examples of the kind of text prompts you can give Fuzz and the music it will generate. The samples are nearly three minutes long, which is on par with what you’d get from a song by a band. But, again, the most impressive experience is available if you go to the Riffusion site.
You don’t have to create an account to listen to music. Plenty of AI song samples are already there, in full length, organized into playlists. As I write this article, I’ve moved through various playlists, and each one impressed me.
I didn’t even bother with the music generation part, which requires registering an account before you’re able to make your own songs. A few months ago, I used ChatGPT to create a specific Apple Music playlist for my marathon running needs, starting with a Taylor Swift track. This was genuine music from real artists, of course.
Fuzz might help me make my own music that sticks to the strict parameters and style I want, and I’d probably enjoy it just as much as music created by human artists.
As VentureBeat explains, Fuzz can also generate complete songs from audio clips and visual prompts, not just text descriptions.
Also impressive is the AI’s ability to learn your music preferences over time and offer personalized music tracks.
Riffusion is a small AI startup, but the company had a brilliant idea of using image diffusion technology to create music. VentureBeat notes that tech giants, including Google, Meta, and TikTok, have all looked at Riffusion’s idea for their own AI music generation apps.
Fuzz is so good right now that it outperformed competitive models in blind human evaluations. That’s impressive, considering the startup doesn’t have the massive resources that big AI firms do.
You don’t have to take my word for it. Just head over to Riffusion and start listening to AI-generated tracks. You can then create your own music and stream everything for free. Hopefully, those GPUs will hold for as long as possible, but Fuzz AI might just be the kind of AI I would consider paying a subscription for.
