Developers can create character voices and localized narration while choosing a model size suited to their serving environment. Core capabilities include Voice design, Reference-voice cloning, Director-style instructions, and Paralinguistic tags.
Public code or model weights are linked from the project, with setup instructions for running or adapting the system. Additional characteristics include Seven languages and two dialects, A 0.5B dense model, A 7.9B-total MoE variant, and Podcast generation in the tiny variant.
