ByteDance built Seed Audio as a fundamentally different kind of audio generation model. It is not just another text-to-speech engine. Seed Audio 1.0 produces expressive, natural-sounding speech and audio scenes including radio dramas, narration, dialogue, and rich ambient soundscapes from a single text prompt.
It is now live in Kolbo's Text-to-Sound tool.
What Seed Audio 1.0 Does
Natural, expressive speech. The output sounds like a real person speaking with purpose, not a synthetic reader moving through words. Inflection, pacing, and emphasis come through without manual tuning.
Preset voices. A library of distinct voices ships with the model, each with its own character. Pick the voice that fits your project and go.
Voice cloning from a reference clip. Provide up to 3 short audio samples and Seed Audio learns the speaker's unique voice characteristics, then generates new speech in that voice. No training run, no upload queue: the clone is active immediately.
Reference image support. Attach an image alongside your prompt and the audio generation takes visual context into account. Useful for tying narration to specific scenes or character contexts.
Fine-grained control. Adjust speed, volume, and pitch as independent parameters. Slow a delivery for dramatic narration, raise pitch for a lighter character voice, or compress the dynamic range for broadcast use.
Output format: MP3. Clean, ready to drop into your edit.
What You Can Make With It
Seed Audio 1.0 is particularly well-suited for:
- Narration and voiceover for video projects, product demos, and explainers
- Character dialogue for animated or AI-generated videos
- Radio drama and audio storytelling with multiple cloned voices in the same session
- Accessibility audio for documents and interfaces
- Podcast intros and outros with a consistent branded voice
How to Try It
Seed Audio is already in your Kolbo workspace.
- Open Audio Tools from your dashboard
- Select Text to Sound
- In the model selector, choose Seed Audio
- Write your text, pick a voice, and generate
To use voice cloning, upload 1 to 3 short audio samples in the reference section and the model will match the speaker's voice.
Seed Audio 1.0 is live in your Kolbo workspace.
Try Seed Audio 1.0Best, Zohar Founder, Kolbo.AI
