On Craze, Seed Audio 1.0 generates speech from a script. You can optionally supply audio or image references, but not both types together.

Why creators choose Seed Audio 1.0

Seed Audio 1.0 accepts references as well as text, giving you another way to guide narration.

Script-to-speech in one generation

Write the line you want spoken. Craze sends that script to the text-to-speech route and returns an audio file.

Reference-guided voice cloning

Upload up to three audio clips, each at most 30 seconds, to guide the generated voice. Alternatively, provide one image reference. You cannot mix audio and image references in one generation.

Multi-character dialogue in one pass

For different characters, generate lines separately and assemble them in the editor. The Craze route does not expose a multi-speaker assignment control.

Precise timing control

Place the generated speech in the editor and adjust its timing against your video there. This route does not expose millisecond-level timing annotations.

Multilingual support with voice conversion

Provide your narration script in the language you want to generate. For a localized version, submit a translated script as another generation.

How to use Seed Audio 1.0 in Craze

1. Write the narration script, up to 2,048 characters.
2. Optionally upload up to three audio references of at most 30 seconds each, or one image reference.
3. Name Seed Audio 1.0 in chat or select it in the Models panel.
4. Craze generates speech and places the audio in your project.
5. Trim, adjust, or regenerate if you want a different take.

Seed Audio 1.0 pricing

Pricing starts at $0.264 per 1,000 characters and varies with script length. The rate card has no reference-type price axis. The calculator shows the total before you generate.

Who Seed Audio 1.0 is for

Podcast and audio drama production

Generate narration for a podcast or audio drama, then add other voices, music, and ambience as separate tracks in your edit.

Branded audio content and ads

Generate the voiceover for an ad and add music or sound effects separately in your edit.

Game dialogue with foley

Generate character lines separately, then layer ambience and foley in the editor.

Localization for global reach

Create localized narration by submitting a translated script for each language.

Seed Audio 1.0 vs Other Models

MiniMax Speech 2.8 HD is text-to-speech with optional voice ID and sound tags. Seed Audio 1.0 also generates speech but accepts audio or image references. Neither Craze route is a complete music-and-foley mixer.

For expressive English narration, Gemini 3.1 Flash TTS offers directorial control. For multilingual narration with emotion tags, Qwen Audio 3.0 TTS is another option. Seed Speech offers text narration with optional voice ID.

Reference Audio Guidance

Seed Audio 1.0 accepts up to three audio references, each at most 30 seconds, or one image reference. Choose one reference type per request. The route does not expose speaker-to-reference assignments.

Music and Foley Generation

The Craze route is for speech. For a coffee-shop scene, generate the barista's line here, then add music and ambience with separate tools.

Edit the speech track's timing and level alongside separately generated sound in your project.

For foley such as footsteps or a door opening, use a sound-effects model and mix the result with your narration.

Multi-Character Dialogue Workflow

For two speakers, generate their lines separately with the appropriate references and arrange the audio in the editor. Craze does not map multiple references to named speakers in one TTS request.

Keep each script within 2,048 characters. Assemble longer exchanges from multiple generated clips.

Timing Control

This Craze route does not expose a timing-annotation control. Adjust the generated speech's position in your edit to match a visual beat.

Multilingual and Localization

Write the translated line as the script for each localized generation. An optional reference can guide the voice, but this Craze route does not offer a separate voice-conversion task.