Seth Forsgren, Co-Founder and CEO of Riffusion, leads the development of an AI-powered music generation tool that creates audio tracks from text prompts. Riffusion allows users to experiment with different musical styles and sounds in real-time, making music creation more accessible. Designed for both creativity and ease of use, the platform enables anyone to explore AI-generated music without requiring formal musical expertise.
Riffusion is the best AI music generator that I have personally tried, and is a tool that I would recommend for users who are interested in AI generated music.
Can you take us back to the early days of Riffusion? What was the initial spark that led you to build an AI music generation tool?
Riffusion started as a total hobby project between two lifelong musicians. My co-founder, Hayk, and I have been playing in amateur bands together for over a decade, and we’ve always been fascinated by the creative act. One afternoon we were in my backyard writing a song, and in the search for inspiration we started playing with early AI models that could generate images out of nothing. But what we really wanted was a tool that could make music with us, an AI we could collaborate with to conjure new melodies and sounds that no one had ever heard. There was nothing like it at the time, so we tried to build it for ourselves more than anyone else.
At what moment did you realize it had the potential to become a full-fledged company?
The turning point came when we shared our hobby project with a few friends, and out of nowhere it went crazy viral. It wasn’t just technologists or AI enthusiasts who were interested – professional musicians, producers, and millions of everyday music lovers were engaging with it in ways we could have never anticipated. Some of our favorite artists in the world started reaching out with samples they had created using Riffusion!
The project also inspired top research labs at Google, ByteDance, and others to spin up their own AI music efforts based on our work, and it was clear that this wasn’t just an experiment – it was the foundation for something much bigger. As a company, we now have the chance to bring this new instrument to creative people everywhere.
What were the biggest technical and business challenges you faced when transitioning Riffusion from an experiment into a commercial product?
On the technical side, we’ve come a long way. Our first models generated grainy, five-second snippets of low fidelity music, and we can now generate full-length, high-quality songs with great controllability and expressiveness. This has taken major advances in the model architecture and constantly rethinking things from scratch. It’s a credit to the amazing researchers on our team that we’ve come this far, and we know this is still just the beginning of what is possible with the tech.
On the business side, we had to think deeply about where Riffusion fits within the music industry. AI music is still new, and while we’re seeing incredible adoption from amateur creators and professionals alike, there’s an ongoing conversation about how AI and human creativity can coexist. Our focus has always been on empowering musicians, not replacing them – giving people new tools to explore their creativity in ways they never thought possible.
Riffusion originally focused on generating short musical riffs, but now it can compose full-length pieces. What advancements allowed you to extend its capabilities?
By training our own foundation models from scratch, we’ve been able to improve the quality, expressiveness, and controllability of Riffusion’s output. A major breakthrough came with the development and release of our latest model, FUZZ. In blind testing, FUZZ consistently outperforms competing models when given the same lyrics and sound prompts, and the model is uniquely designed to help users find their personal sound – the more time a user spends using Riffusion, the more FUZZ learns their individual taste, and the more personalized the generated music becomes. We see this as a big differentiator for Riffusion.
Many AI music models struggle with maintaining emotional depth in music. How does Riffusion capture the nuances of different moods and styles?
Music is deeply personal and emotional, and we want Riffusion to generate music that resonates on a human level. As our advisor Alex Pall says, “This isn’t about making sound; it’s about making people feel something special through sound.”
Just like a well-made violin can empower an artist to express themselves fully, we train our models to be an instrument guided by the user’s creativity at every turn. Whether you’re inputting a melody, a text prompt, or even an image, Riffusion adapts to your intent, shaping the output to reflect different moods, dynamics, and stylistic choices. We’re focused on the users who come back day after day to craft really amazing music on the platform.
As AI-generated music continues to evolve, how do you see it complementing rather than replacing human creativity?
AI is an instrument for musicians, not a replacement. Throughout history, new musical tools – from synthesizers to digital audio workstations – have expanded what artists can create without diminishing human artistry. Riffusion follows that same philosophy. We see Riffusion as an instrument that encourages musicians to experiment, collaborate, and try new forms of storytelling. Artists still bring the soul and intent to their music, and AI helps bring those ideas to life. We’re thrilled that so many professional musicians and enthusiasts are embracing this tool every day, and finding joy in the creative process.
How have musicians and producers responded to Riffusion’s capabilities? Have you seen any unexpected or innovative uses of the tool?
The response has been pretty incredible. There were a few who started using these tools years ago, but every week now the number of professional musicians and producers who are embracing the technology is expanding dramatically. We’ve seen artists use Riffusion for brainstorming new melodies, crafting new sounds, and even composing entire albums. Some have blended Riffusion’s generations with live instrumentation to create entirely new genres. One of the most exciting things has been seeing how people take this tool and make it their own, whether it’s generating music from the sounds of nature, testing experimental compositions, or scoring cinematic pieces.
Now that Riffusion can create long-form music, do you see potential for AI-generated scores in film, video games, or other media?
The ability to generate long-form music certainly makes Riffusion a powerful tool for the larger media landscape. We’re already seeing interest from filmmakers, game developers, and content creators who want unique scores that adapt to their narratives, perhaps even in real-time. It’s clear that AI can help storytellers express themselves across modalities, and we’re just at the start of this frontier.
Looking ahead, what is your ultimate vision for Riffusion, and how do you see it shaping the future of music creation?
Today, only a tiny fraction of people make music, but creativity is inherent in all of us. By building tools that lower the barrier to entry while also raising the ceiling of what’s possible with sound, Riffusion will be an important instrument in the future. I see Riffusion as the go-to place for anyone looking to create, experiment, and connect through music. Whether you’re a professional producer or someone who’s never made a song before, we want Riffusion to be the instrument that helps you find your sound.
Thank you for the great interview, users who are interested in generating some music should visit Riffusion.
Credit: Source link
