Seed Audio 2.0 is an upcoming audio generation model for creating longer, multi-part scenes with dialogue, music, ambience, and effects. It is aimed at film and comic-drama teams, advertisers, game audio teams, and localization producers who need editable audio parts or video-aware dubbing.

Create scenes up to six minutes long with up to six reference voices.
Generate from text, reference audio, video, or combinations of those inputs.
Produce separate dialogue, music, ambience, and effects stems for editing.
Use video context for dubbing that follows action, cuts, and pacing.
Work with a published specification covering 30 languages and timestamped cues.

Unlike a single mixed audio output, the planned model specifies independent stems and picture-aware generation for production workflows. Seed Audio 2.0 is marked as coming soon; the live studio currently generates with Seed Audio 1.0, where prompts and references can be rehearsed before the new model is enabled.