MiniMax Voice Design is available on Novita AI as a pay-per-voice API. You describe a voice in plain text — tone, gender, accent, energy — and the model returns a voice_id you can drop straight into any MiniMax Speech 2.8 TTS call. No reference audio, no recording session, no manual voice editing. If you’re building characters for a game, diverse voices for a content pipeline, or a consistent brand voice for an agent, this is the fastest path from description to deployable voice.
Key Takeaways
- Voice Design generates a voice from a text prompt, not from a reference audio file — a different approach from voice cloning.
- The API returns a
voice_idthat works with MiniMax Speech 2.8 HD and Turbo endpoints on Novita AI. - Designed voice IDs are temporary (7-day expiry) unless activated by a speech synthesis call, after which they persist permanently.
- Preview audio generation is billed separately from the voice creation fee — budget for both.
- Voice Design costs more per voice than Quick Voice Cloning; use it only when you need a voice that doesn’t exist yet.
What Is MiniMax Voice Design?
MiniMax Voice Design is a voice generation model that accepts a natural language description and produces a synthetic AI voice matched to those characteristics. You specify what you want — “a calm, authoritative female narrator with a British accent” or “an excited young male tech reviewer, fast-paced and high-energy” — and the model synthesizes a voice from that description.
The output is a voice_id string, not a finished audio clip. That voice ID is what you then pass to the TTS endpoint to generate actual speech from any text input. The model also generates a short preview audio sample during the design step so you can evaluate the voice before committing it to production.
This is fundamentally different from voice cloning, where you upload an existing recording and the model replicates it. Voice Design creates a voice that didn’t exist before. That distinction matters for content that needs original, non-attributable voices: fictional characters, brand assistants, synthesized narrators, NPC dialogue.
MiniMax Voice Design API Access on Novita AI
MiniMax Voice Design is available at novita.ai/models/minimax/minimax-voice-design. You can try it in the Novita playground without writing code first.
For API access, the endpoint follows Novita’s standard model API pattern:
POST https://api.novita.ai/v3/minimax-voice-design
Authorization: Bearer YOUR_NOVITA_API_KEYThe request body takes two fields: prompt (your voice description) and preview_text (the sample sentence the designed voice will speak in the preview audio):
{
"prompt": "Excited and enthusiastic male product reviewer, fast-paced, high energy, persuasive.",
"preview_text": "What is up, everyone! Today we are unboxing the brand new Gadget X-Pro, and I have to say — this thing is incredible."
}The response includes a voice_id string and the preview audio encoded as hex. The voice_id is what you’ll pass to Speech 2.8 endpoints for actual synthesis.
Get your API key from the Novita AI console. New accounts receive free credits.
Specs and Pricing Summary
| Field | Details |
|---|---|
| Display name | MiniMax Voice Design |
| Model ID | minimax-voice-design |
| Base URL | https://api.novita.ai/v3 |
| Endpoint | POST /v3/minimax-voice-design |
| Voice creation fee | ~$3.00 / voice (verify at Novita console) |
| Preview audio fee | ~$30 / 1M characters generated during preview |
| Voice ID expiry | 7 days if unused; permanent once used in TTS |
| Downstream TTS pricing | $60 / 1M chars (Turbo) or $100 / 1M chars (HD) |
| Best fit | Original voice creation for characters, brands, pipelines |
Pricing sourced from MiniMax platform docs and Novita published articles, Jul 2026; verify current Novita rates at novita.ai/console/pricing-console.
Key Capabilities for Developers
Prompt-driven voice synthesis. The voice description is freeform text. You can specify gender, age range, accent, speaking pace, energy level, tone, and emotional quality. More specific prompts generally yield more distinctive voices; vague prompts produce a generic output.
Reusable voice IDs. Once you have a voice_id, you can reuse it across as many TTS calls as needed. A single voice creation fee covers unlimited downstream speech synthesis using that ID.
Persistence model. Voice IDs are temporary by default and expire after 7 days. Making at least one TTS synthesis call using the voice_id converts it to a permanent voice. If you’re building a production pipeline, activate the voice ID before the 7-day window closes.
Preview before commit. The voice design step returns a preview audio sample alongside the voice_id. Evaluate the voice before using it in production speech synthesis — iterate on the prompt if needed.
Multilingual output. Designed voices work with MiniMax Speech 2.8’s 40+ language support. A voice designed with an English-language prompt can synthesize speech in other languages while preserving the tonal characteristics.
Compatible with Speech 2.8 features. Designed voice IDs work with voice mixing (timber_weights), emotion presets, and tone tags like (laughs) and (sighs) available in the Speech 2.8 series.
When to Use MiniMax Voice Design
You need a voice that doesn’t exist. If you’re building a fictional character, an AI brand persona, or NPC dialogue where there’s no real person to clone, Voice Design is the right starting point. Describe the voice you want and iterate on the prompt until the preview sounds right.
You’re scaling a content pipeline with diverse voices. A documentary series, an e-learning platform, or a localization project that needs dozens of distinct, original voices can use Voice Design to create a library of voice IDs in minutes. Each voice ID is then reusable without additional creation fees.
You want to avoid rights and consent overhead. Voice cloning requires a real voice recording, which carries consent and rights considerations. Voice Design produces an entirely synthetic voice with no real-world voice identity behind it.
Your use case pairs naturally with Speech 2.8 expressive features. If you’re planning to use emotional tone tags or emotion presets in your TTS calls, Voice Design can create voices specifically tuned to those emotional ranges.
When Not to Use MiniMax Voice Design
You need to replicate a specific person’s voice. If you have a real reference recording and want to match it, MiniMax Quick Voice Cloning on Novita AI is the right tool. Cloning is also cheaper per voice at approximately $1.50 vs. $3.00 for Voice Design.
You need system voices. If you don’t require a custom voice, Novita AI’s Speech 2.8 endpoints include a library of built-in voices at no creation cost. Use those for generic narration, notifications, or any scenario where voice identity isn’t a differentiator.
You’re running a quick prototype. The voice creation and preview step adds latency to the development cycle. For a prototype where voice quality doesn’t matter yet, a system voice is faster to iterate with.
You need tight latency on voice creation. Voice Design is a synchronous one-shot call, but iterating on voice prompts to get the right output takes time. If you’re building a user-facing workflow where the end user designs their own voice in real-time, plan for iteration cycles.
How It Fits Your API Workflow
Voice Design is the front step in a two-step voice pipeline:
- Design step — Call
POST /v3/minimax-voice-designwith a prompt and preview text. Evaluate the returned preview audio. Adjust the prompt and repeat if needed. - Synthesis step — Once satisfied, pass the returned
voice_idtoPOST /v3/minimax-speech-2.8-hdorPOST /v3/minimax-speech-2.8-turboas thevoice_idparameter.
The speech synthesis call activates and permanently saves the voice ID. From that point, the voice is available indefinitely.
For production pipelines, run the design step offline during setup. Store the resulting voice_id values in your configuration layer. Runtime TTS calls then only hit the synthesis endpoint — fast, predictable, and no per-call voice creation cost.
If you want to expose voice customization to end users (for example, a character creation flow in a game), run Voice Design on demand when the user finalizes their choice, then activate it with an immediate TTS call to lock in the voice ID.
For more on the downstream TTS workflow with Speech 2.8, see MiniMax Speech 2.8 Series on Novita AI.
Conclusion
MiniMax Voice Design is worth it if you need original voices that don’t exist in any reference recording. The $3.00-per-voice creation fee is a reasonable one-time cost for a reusable voice ID across a full content pipeline. The critical constraint to know upfront is the 7-day expiry — activate every designed voice with a synthesis call before it expires, or you’ll pay to recreate it.
For any scenario where a real recording exists, cloning is cheaper and more predictable. For anything that just needs a working voice, the built-in system voices are free. Voice Design sits in a specific niche between those two options — original, controllable, and scalable.
Get started at novita.ai/models/minimax/minimax-voice-design. Free credits apply on new accounts.
FAQ
What is MiniMax Voice Design?
MiniMax Voice Design is an API that generates a synthetic AI voice from a text prompt describing its characteristics. It returns a voice_id for use in downstream text-to-speech calls — no recording or reference audio is required.
How does Voice Design differ from Voice Cloning? Voice Cloning replicates an existing voice from an uploaded audio sample. Voice Design creates an entirely new voice from a text description. Cloning matches a specific real voice; Design creates an original one.
How long does a designed voice ID last?
Voice IDs expire after 7 days if unused. Making at least one TTS synthesis call using the voice_id makes it permanent.
Can I use a designed voice with Speech 2.8 features like tone tags?
Yes. Designed voice_id values work with all MiniMax Speech 2.8 capabilities on Novita AI, including emotional tone tags ((laughs), (sighs)), emotion presets, voice mixing, and multi-language output.
What does MiniMax Voice Design cost on Novita AI? Voice creation is approximately $3.00 per voice (matching MiniMax’s official platform rate). Preview audio generation during the design step is billed at approximately $30 per 1M characters. Check current Novita AI pricing at novita.ai/console/pricing-console before production use.
How do I try MiniMax Voice Design without code? Go to novita.ai/models/minimax/minimax-voice-design and use the playground. New accounts get free credits.
