Output fine-tuned Stable Audio on its own sample library to power the Co-Producer tool, giving musicians endless variations in seconds.
At a glance: Why Output uses Stable Audio
| Company | Output is a global leader in music creation software, featured on tracks from Grammy winning artists like Beyoncé, Kendrick Lamar, and Billie Eilish. |
|---|---|
| Use case | Creating sample variations in Co-Producer, a plugin that helps you find or create premium, royalty-free samples that fit right into your music |
| Models used | Stable Audio Open (2025 launch), and now Stable Audio 3 Medium |
| Key capabilities |
|
Building Co-Producer: From finding sounds to creating new ones
Output is a global leader in music creation software, powering the world's biggest hitmakers. Best known for its groundbreaking creative plugins and instruments like Arcade, Exhale, Rev, and Portal, Output builds tools that push the boundaries of musical creativity without replacing the human at the center.
Output’s Co-Producer plugin helps producers find or create samples by putting Co-Producer on their master track and allowing it to listen to what they’re working on, such as a guitar part or a track in progress. Co-Producer listens to it and suggests samples from Output's and other third party libraries that fit, like drums, keys, or bass.
The first version of Co-Producer used AI to listen to a musician’s audio, analyze it, and recommend existing samples from Output’s library.
The new Re-imagine feature was an opportunity to go beyond recommending sounds to generating them. Once a musician found a promising sample, what if they could explore endless new versions of it? But Output needed to find the right AI tool.
Most AI tools don’t fit how music really gets made
Output has spent over a decade building tools for musicians and producers. The team knows how artists actually work…and where most AI tools miss.
"If you look at some generative music and AI music products out there, they're very disconnected from how people make music today," says Spencer Salazar, CTO at Output. "We understand how people make music, what their needs and preferences are, and what the different types of creators are."
Musicians and producers work in iterative elements: a drum loop, a keys phrase, a texture to layer against their own playing. Many of the tools on the market went the other way, generating fully mixed tracks, often with vocals on top. Output wanted AI that respected the way its users already work.
"No one who is serious about music making wants to type in a prompt and make an entire song, and then that's the entire extent of their musical journey," he says. "People want to feel like they're part of the process, not a bystander.”
Output’s approach: Customizing Stable Audio for distinctive sounds and fast iteration
In early 2025, Output's team downloaded the open weights of Stable Audio Open and began experimenting with fine-tuning. Output already had a deep internal sample library; custom fine-tuning provided a way to adapt the model toward that material and the kinds of sounds its customers wanted to use.
When the results showed promise for a customer-facing feature, Output licensed the model for commercial use as an enterprise. All Stable Audio models are trained on fully licensed material, and you can commercialize outputs.
Later, Output moved its fine-tuning process to Stable Audio 3.0, the next-gen model architecture. Today, a custom fine-tune of Stable Audio 3.0 Medium is the generative audio engine that powers Co-Producer.
Customization: Variations that sound like your library
Fine-tuning gives businesses a way to differentiate their products using material they already have, Salazar says. That means they can offer something more distinctive than the same general-purpose generation available everywhere else.
"If you have a lot of drum and bass samples, you can make the best drum and bass sample generator there is," he says.
He sees similar potential for individual artists with collections of recordings and unreleased material. Fine-tuning could let them explore new sounds informed by their own work, creating a more personal starting point for experimentation.
Fast generation: An experience that responds like an instrument
Musical quality was only one of the requirements. The experience also needed to feel fast and responsive enough for musicians to keep experimenting.
Salazar compares a good tool to a guitar: touch it and it makes a sound, instantly. That reactivity is what makes an instrument learnable, because players build a mental map of action and result. Output applied the same standard to Co-Producer. It generates short snippets, optimized so producers get results in under five seconds and can keep playing instead of waiting.
Stable Audio 3.0 has industry-leading inference speeds, with sub-second generation on self-hosted GPUs, and a few seconds via API.
"That's a huge part of creating tools that musicians actually enjoy using," he says. "They're responsive and learnable.”
What it’s like working with the Stable Audio team
Output evaluated other audio model providers before choosing Stable Audio. A few things set the partnership apart.
Flexible experimentation. Output could download an open-weights model, fine-tune it, and test it on its own terms before committing. Elsewhere, the fine-tuning trial had limitations that made evaluating the results challenging, such as the number of songs to fine-tune on.
A collaborative partnership. Output shares Stable Audio’s artist-first approach to building tools. "It feels like more of a partnership," Salazar says. "The team understands musicians, and we're building toward the same goal: tools that artists actually want to use."
Openness with the research community. Salazar, who comes from academic research, regularly sees Stability AI researchers at conferences and values their work with student researchers.
What’s the opportunity in generative audio? Adding new possibilities to the creative process.
For companies exploring generative audio, Output's experience shows that the opportunity isn't just to generate audio. It's to make generation useful within the way artists want to work.
Searching for samples has long been part of music production. Output created Co-Producer to help musicians find samples that fit their music. By adding generative audio, Output has given musicians more ways to work with the sounds they find.
"We're not afraid to embrace new technology as long as it's consistent with an underlying set of principles that we share with creative people," Salazar says. “We want to see what kinds of interesting tools new technology can give musicians while still maintaining a high degree of agency over their creative processes.”
Build your next audio capability.
Give your users new ways to create with audio using customizable, commercially safe models, with flexible options for deployment.
