Stable Audio 3's variable-length generation up to six minutes and native inpainting changes the audio production use cases

Variable-length audio generation up to six minutes and twenty seconds without fixed-length padding is the capability that makes Stable Audio 3 useful for long-form content like podcast intro music, background tracks for full videos and extended scene-setting audio. Previous models with fixed-length outputs required looping or manual extension that was audible in the final output.

The native inpainting for audio being available alongside generation is the editing capability that makes Stable Audio 3 useful for refining rather than regenerating. Modifying a specific section of generated audio while preserving the rest of the composition is a production workflow that was not previously available in open-weight audio models.

The open-weight release means self-hosting for commercial use without per-generation API costs is viable for production operations with high audio generation volume.

The distinction between the small, medium and large model variants being relevant for different deployment contexts is worth understanding before choosing which model size fits your specific infrastructure and quality requirements.

For audio producers and music supervisors: at what track length does the six-minute maximum change what Stable Audio 3 is useful for in your production workflow?