HeyGen Voice Turbo: Free TTS Until Nov 30, Explained (2026)
Vorec Team · 2026-10-22 · About 8 min read
HeyGen has opened its own text-to-speech model to developers, and the quicker variant, HeyGen Voice Turbo, can be called through its API at no charge until 30 November 2026. HeyGen also lists two paid options alongside the free one: Turbo at $4 per million characters and the full HeyGen Voice model at $15 per million characters. The free preview has a stated end date; HeyGen says it will give advance notice before anything changes.
For anyone producing narrated tutorials, training videos or product demos, that is worth a look: another voice model is available, on a free trial window with clear limits, and from a company whose main product is video. This post summarizes what HeyGen announced, from HeyGen's announcement and its other own pages (checked 22 October 2026), what "free" covers, and where a tutorial narrator should be careful.
At a glance (from HeyGen's announcement, last updated 21 October 2026): Free model: heygen-voice-1-turbo-free, free through 30 November 2026, best effort. Paid: Turbo $4 and HeyGen Voice $15 per million characters. Voices: public voices and instant clones on Turbo; professional clones need the full model. No credit card required for the free model.
What HeyGen announced
According to HeyGen, HeyGen Voice is the model that produces the speech in all of its videos. Turbo uses the same voices with what HeyGen describes as a faster generation process, which it says makes it quicker and cheaper to run. HeyGen also reports that listeners in blind comparisons rated Turbo slightly above the full model.
The announcement is about developer access. You create a key on HeyGen's API keys page, call its text-to-speech endpoint with the free model name, and get back a link to a WAV file. Moving to production means changing the model name; HeyGen says nothing else in the request changes.
Three HeyGen posts from the past six weeks fit together:
| Date | HeyGen source | What it covers |
|---|---|---|
| September 2026 (monthly update, last updated 1 October) | What's new at HeyGen: September 2026 | Professional Voice Clone available through the API |
| 14 October 2026 | HeyGen Research technical report | How the HeyGen Voice model is designed and trained |
| Last updated 21 October 2026 | HeyGen Voice Turbo blog post | Free Turbo model for developers, paid prices |
What "free" actually means
HeyGen spells out the terms, and they matter more than the headline:
- Until 30 November 2026. HeyGen says it will give advance notice before anything changes.
- Best effort. There is no uptime or latency guarantee. When free capacity is full, requests return an HTTP 429 error and you retry later.
- Voices. The free model works with HeyGen's public voices and with instant voices you clone from a single recording. Creating instant voices is free during the preview.
- Professional clones are excluded. A professional voice clone needs the paid heygen-voice-1 model.
- No card. A HeyGen account and an API key are enough.
That makes the free model suitable for what HeyGen itself suggests: prototypes and testing. A batch job that has to finish before a course launch should not rely on a best-effort endpoint that can return 429 at busy times.
The paid prices, and what they mean per tutorial
HeyGen's table lists three options:
| Model | Price | Speed | Voices | HeyGen's suggested use |
|---|---|---|---|---|
| Turbo Free | $0 until 30 Nov 2026 | Best effort | Public and instant | Prototypes and testing |
| Turbo | $4 per 1M characters | Fast, priority | Public and instant | Production at volume |
| HeyGen Voice | $15 per 1M characters | Standard | Public, instant and professional | Your own trained voice |
To make those numbers concrete, here is our own worked example. The inputs are assumptions, not HeyGen figures: a 10-minute tutorial narrated at about 150 words per minute is roughly 1,500 words, and at an average of about 6 characters per word including spaces, that is about 9,000 characters.
- Turbo: 9,000 / 1,000,000 x $4 = $0.036
- HeyGen Voice: 9,000 / 1,000,000 x $15 = $0.135
At those rates, character costs are small next to the time spent writing and checking the script. Re-generating narration after every script edit is what multiplies the character count, so batch your edits.
The research behind it: scene-aware voice cloning
A technical report from HeyGen Research (14 October 2026) explains what makes the model different from a plain text-to-speech system. HeyGen Voice takes a recording of your voice, a script and a target image, and produces speech meant to fit the person and place on screen.
In HeyGen's description:
- The voice reference establishes who is speaking; the image and scene descriptions guide how the speech should sound.
- A language model turns the image and requested delivery style into two text descriptions: one for the speaker (tone, pace, emphasis, expressiveness) and one for the background (the space and its sound, such as a close microphone in a quiet studio, or traffic).
- The model generates 44.1 kHz audio.
- For longer scripts, it reuses an audio anchor to carry the same voice, delivery and acoustics across segments, and each segment receives the end of the previous speech as context.
The report's demos compare HeyGen Voice with other voice models in scenes such as a beach, a moving car and a factory floor, all rendered with HeyGen's own avatar video model.
Note the difference between the research and the API. The report describes the model's design, including image conditioning and the audio anchor. The free API announcement documents a simpler request: a model name, a voice and the text. We did not test the API, and nothing in the announcement says the image or scene controls are exposed in the free endpoint.
About the leaderboard claim
HeyGen states that Turbo ranks first and HeyGen Voice second on the Artificial Analysis text-to-speech leaderboard, in a controlled comparison where every model reads the same eight cloned voices (four US, four UK) and listeners pick the better of two clips. HeyGen's page says it captured the ranking on 20 October 2026. We are reporting HeyGen's statement; we did not check the leaderboard ourselves, and a controlled cloned-voice comparison tells you about that test, not about every voice, language or use case. Leaderboards also move as new models arrive.
What this means for tutorial narration
The following is our reading for people who make screen tutorials, training videos and product demos. It is not something HeyGen claims.
1. Scene matching matters less when nobody is on screen. The research is built for talking videos, where a voice has to sound like it belongs in the shot. A screen tutorial usually has no presenter in frame, and the expected sound is a clean, close-microphone narration. The parts of the design most relevant to tutorials are elsewhere: consistent voice across a long script, and clear speech.
2. Consistency across segments is the feature to test. In tutorials generated in segments, one per step, a voice that drifts between step 3 and step 4 is distracting. HeyGen's audio-anchor approach targets exactly that. Test it with your own multi-step script before relying on it.
3. One instructor voice across a course. HeyGen's September update gives this example directly: with Professional Voice Clone in the API, a training team can keep every course in the same instructor's voice. Per that September post, a professional clone is trained from at least 20 minutes of one speaker's recordings, each voice needs a voice slot purchased on the API usage page (each slot includes five trainings per billing period), and synthesis bills 0.6 API credits per generated minute. That per-minute credit figure comes from the September post and the per-character prices from the October one; they are separate statements, so check current billing for the model you call before you budget. Cloning a real person's voice needs that person's informed consent; our guide to AI voice cloning and consent covers what to check.
4. The voice is the last step, not the first. A better voice does not fix narration that describes the wrong thing or runs ahead of the screen. The script has to match the actions, and the timing has to fit. Our tutorial video script guide covers the first part.
Where Vorec fits
Vorec handles capture, narration drafting and timing as well as the voiceover; this article makes no claim about a HeyGen integration, and we haven't tested one. It records your screen with its own macOS recorder (an AI agent can drive it through the Claude Code plugin, capturing locally for you to review before anything is uploaded), or you can upload an existing recording. It drafts narration matched to the captured workflow and generates the voiceover, so nothing is spoken into a microphone, and freeze-sync can hold a frame to give an explanation time to finish. Narration regenerates in supported languages on eligible plans, and the same capture can also produce a written step-by-step guide.
FAQ
Is HeyGen Voice Turbo free?
Through the HeyGen API, the heygen-voice-1-turbo-free model is free until 30 November 2026, on a best-effort basis with no uptime or latency guarantee. HeyGen says it will give advance notice before anything changes.
How much does HeyGen Voice cost after the free period?
HeyGen lists Turbo at $4 per million characters and HeyGen Voice at $15 per million characters.
Can I use my own voice?
Yes. HeyGen says you can clone an instant voice from one recording and use it with Turbo. Professional voice clones, trained from at least 20 minutes of recordings and needing a purchased voice slot, use the full heygen-voice-1 model.
Do I need a credit card for the free model?
No, according to HeyGen: an account and an API key are enough.
Is HeyGen Voice the best text-to-speech model?
HeyGen reports first and second place on one controlled leaderboard as of 20 October 2026. That is one test, run on cloned voices; compare voices on your own scripts and languages before choosing.
How does this relate to HeyGen's video plans?
This announcement covers API access to the voice model. For HeyGen's video plans and credits, see our HeyGen pricing breakdown.
Related reading: HeyGen alternatives for avatars and tutorials, AI voiceover for videos, Eleven v4 and OpenAI prompt-to-voice.
Sources (all HeyGen's own pages, checked 22 October 2026): HeyGen Voice Turbo: now free in your API (last updated 21 October 2026); HeyGen Research, HeyGen Voice: Scene-Aware Voice Cloning for Talking Videos (14 October 2026); What's new at HeyGen: September 2026 (last updated 1 October 2026).
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. Trial includes up to 3 projects; exports carry a watermark.