OpenAI Prompt-to-Voice: What It Means for Narration (2026)
Vorec Team · 2026-10-17 · About 7 min read
Picking a narrator voice for a tutorial series usually means scrolling through a voice library until something sounds close enough. On 16 October 2026 OpenAI shipped a different approach for API customers. You describe the voice you want in plain language, for example "a warm adult voice with a mid-range pitch, unhurried delivery", and the API saves a custom voice you can reuse.
The feature is called prompt-to-voice. It is useful to know about. There is also one limit you need before you plan a narration workflow around it: according to OpenAI's own guide, prompt-generated voices work with GPT-Live, OpenAI's real-time voice conversation API. They aren't supported by the Text to Speech API.
This post covers what OpenAI published, the limits it states, and what that means for people who narrate software tutorials and product walkthroughs. We have not tested the feature. Where something is our reading rather than OpenAI's statement, we say so.
Checked on 17 October 2026. Sources: the OpenAI API changelog entry dated 16 October 2026, and OpenAI's Custom voices guide. Both are OpenAI's own documentation. API products change often, so re-check them before you build on this.
What OpenAI released
The 16 October changelog entry says OpenAI:
- Released prompt-to-voice for all API customers. You create a custom voice from a text description through the API or the voice Playground, then reuse it in GPT-Live sessions.
- Made creating a prompt-generated voice free. Speech generation is billed at regular GPT-Live session rates.
- Added nine natural voices for GPT-Live in German, French, Hindi, Japanese and Korean.
The Custom voices guide adds the details that matter in practice:
| Point | What the guide says |
|---|---|
| Where the voice works | GPT-Live. "They aren't supported by the Text to Speech, Realtime, or Chat Completions APIs." |
| What you send | A JSON request to `/v1/audio/voices` with `type` set to `prompt`, a name and a description |
| Optional script | `script_hint`, up to 2,000 characters of representative speech; if omitted, the service generates one |
| What you get back | Metadata: an ID, a creation time and the original prompt. No audio and no preview |
| Limit | Up to 100 prompt-generated voices per organization, including voices your app creates for its customers |
| Consent recording | Not needed for prompt-to-voice |
| Cost to create | Free; listening in the Playground runs a GPT-Live session and is billed |
The guide also lists the qualities you can describe in the prompt: accent, age, presentation, pitch, timbre, texture, loudness, speech rate and distinctiveness. These are words in your prompt, not separate API parameters.
The rules OpenAI attaches
The guide sets out responsible-use rules that apply to anyone using prompt-generated voices:
- No impersonation. Don't ask the model to imitate a named person, public figure or fictional character. Describe the qualities themselves.
- Adult voices only. Children's voices aren't allowed.
- Be transparent. Tell listeners the speech is AI-generated, and preserve any provenance metadata, watermarks or other indicators of synthetic origin.
- No uniqueness guarantee. OpenAI says others may generate similar voices.
The second path in the same guide, custom voices from recordings, works differently. It needs a consent recording, in which the voice actor reads a fixed consent phrase, plus a reference sample of 30 seconds or less. The guide says those voices are limited to eligible customers and capped at 20 per organization. Unlike prompt-generated voices, they can be used with the Text to Speech API, the Realtime API and Chat Completions audio output, as well as GPT-Live.
What this means for tutorial narration
What follows is our reading of the documentation, not a statement from OpenAI.
It is not a drop-in text-to-speech voice. A narrated tutorial is usually a file: a script is turned into audio, timed against the video and exported. That is normally a text-to-speech job. Because prompt-generated voices are not supported by the Text to Speech API, you can't swap one into a standard "script in, audio file out" call today. GPT-Live is a full-duplex conversation API. A custom pipeline that captures GPT-Live audio for a recorded video may be possible, but the guide we read doesn't describe that use and we haven't tested it. What the guide does describe is using the voice in GPT-Live sessions, which suits live voice experiences such as an assistant.
Describing a voice is still a useful habit. The guide's list of qualities (pitch, pace, timbre, vocal effort) is a good way to write a voice brief whatever tool you use. A one-line brief like "calm adult voice, mid-range pitch, medium pace, moderate vocal effort" makes it easier to pick a consistent narrator from any voice library, and to explain the choice to a teammate.
Keep the prompt. OpenAI recommends saving the voice ID and the original prompt, and including the description in session instructions for voice performance. For a tutorial series, the equivalent is to document the voice you chose, so videos made months apart sound like one series.
Disclosure is part of the job. OpenAI's rules require telling listeners the speech is AI-generated. Whatever provider you use, follow its disclosure requirements for the context where the video will be shown, and check any rules that apply where you publish.
A conversational voice is not a timed voice. In a tutorial, the voice has to land on the click it describes. That timing comes from the script and the editor, not from the voice model. A great voice reading a line that runs past the action on screen still produces a confusing tutorial.
How to choose a narrator voice, whatever the tool
- Write the brief first. Describe accent, pitch, pace and energy in one sentence before you listen to anything.
- Test with a real script. Use a paragraph with your product name, menu labels and a keyboard shortcut. Sample sentences hide the hard cases.
- Listen at tutorial speed. A voice that sounds pleasant in a 5-second preview can feel slow over a 3-minute walkthrough.
- Check every language you ship. If you localise, have a native speaker rate pronunciation and pacing.
- Confirm where the voice can be used. As this release shows, a voice that exists in one API may not be available in another.
- Record the decision. Save the voice name or ID and your brief with the project.
Where Vorec fits
Vorec handles the parts a voice model leaves open: what to say, and when to say it. It records your screen, with its own macOS recorder or from a recording you upload, and drafts narration matched to the workflow it captured, then generates the voiceover, so nothing is spoken into a microphone. In the editor you can add smooth cursor motion, cursor-follow zoom and click-based auto-zoom, and freeze-sync can hold a frame to give an explanation time to finish. Narration regenerates in supported languages on eligible plans without re-recording, and the same capture can also produce a written step-by-step guide.
An AI agent can drive the recording too. Vorec's `record-tutorial` skill lets Claude Code or Codex plan a walkthrough and drive the Vorec recorder through the CLI. The capture stays local for you to review before anything is uploaded and narrated.
For more on voices, see our AI voiceover guide, our explainer on AI voice cloning and consent, and our post on MAI-Voice-2.1, Microsoft's multilingual text-to-speech model. Plans are listed on the Vorec pricing page. Trial includes up to 3 projects; exports carry a watermark.
FAQ
What is OpenAI prompt-to-voice?
A feature of the OpenAI API that creates and saves a custom voice from a text description. OpenAI released it to all API customers on 16 October 2026.
Can I use a prompt-generated voice with the OpenAI Text to Speech API?
Not according to OpenAI's Custom voices guide, checked on 17 October 2026. It says prompt-generated voices work with GPT-Live and aren't supported by the Text to Speech, Realtime or Chat Completions APIs.
How much does prompt-to-voice cost?
OpenAI says creating a prompt-generated voice is free. Generating speech with it is billed at regular GPT-Live session rates, and listening in the Playground runs a billed session.
How many prompt-generated voices can I create?
Up to 100 per organization, according to the guide. The limit also counts voices your application creates for its customers.
Do I need consent to create a prompt-generated voice?
The guide says no consent recording is needed, because the voice comes from a description, not a person's recording. OpenAI's rules still apply: no imitation of named people or characters, adult voices only, and tell listeners the speech is AI-generated.
Is prompt-to-voice the same as voice cloning?
No. Cloning from a recording is a separate path in the same guide. It needs a consent recording and a reference sample, is limited to eligible customers and allows at most 20 such voices per organization.
Want the recording to narrate itself? Record with Vorec, or upload one you already have, and get a narrated tutorial plus a written guide. Start free. 7-day trial, 100 credits, no credit card required. Trial includes up to 3 projects; exports carry a watermark.