Blog

HeyGen Avatar 6: an AI avatar that gestures to your script

HeyGen launched Avatar 6 on November 17, 2026: gestures that follow the script, native 2K, two avatars in one shot. And what its 84.4% figure is really worth.

Frank Houbre

Frank HoubreFounder of Imaginode

4 min read

What HeyGen launched

HeyGen, the company known for talking avatars, launched Avatar 6 on November 17, 2026. The pitch: a digital double that no longer just moves its lips, but acts out the script.

According to the Avatar 6 page, earlier avatars "looked like you but recycled the same gestures". The new model reads the script and picks gestures to match: it points, counts on its fingers, shows a direction. You can also ask for them yourself, along with the emotion and expression you want.

The rest of the list: **native 2K** frames, motion that changes with each line instead of looping, and **two avatars in the same shot** for a host-and-guest conversation generated in one go. According to the press release, the model's context window grows from about 3 to 30 seconds, which leaves room for more varied movement. Avatar 6 is available on the HeyGen platform and through the API, and the page offers to try the two-avatar conversation through MCP.

HeyGen's Avatar 6 page on November 18, 2026.Source: heygen.com

How you build your avatar

The page lists three steps, and the first one matters more than the other two.

You upload a photo, **verify that it is you**, then film yourself talking: at least 15 seconds according to the page's how-to, 30 seconds according to the press release, which also lets you add more footage to vary the gestures. The model picks up your voice, expressions and the way you move. Then you write the script (or upload audio), set the voice and pace, and generate.

On consent, HeyGen writes that only the person themselves, or someone they have clearly given permission to, can create their avatar, after a live identity check. The press release adds an important detail: third-party platforms using the API can generate photo-based or virtual avatars, but realistic videos of real people stay on the HeyGen platform. On price, the page only shows a first video for free, no credit card needed.

HeyGen's press release on PR Newswire, November 17, 2026.Source: prnewswire.com

The 84.4% figure, read closely

HeyGen headlines that Avatar 6 is "84.4% indistinguishable" from real footage. The number is correct on its own terms, but it does not say what most people will think.

The page's FAQ explains it: raters see a real recording and an Avatar 6 clip side by side, saying the same sentence, and pick the real one. They pick the avatar **42.2% of the time** (270 answers). Since 50% would mean nobody can tell, HeyGen doubles the score: 84.4%. In other words, more than half the time, raters still spot the real video.

The published table gives another useful reading: shown alone, a real recording is judged real 78.9% of the time, an Avatar 6 clip with the person's own voice 65.1%, and with a synthetic HeyGen voice 56.5%. **The voice carries a lot of weight.** HeyGen also says 72.4% of raters prefer Avatar 6 to Seedance 2.5. HeyGen ran all of these tests.

The results HeyGen publishes on its Avatar 6 page.Source: heygen.com

What it changes for your videos

The most visible flaw of a talking avatar was never the mouth. It was the body.

An avatar that repeats the same shrug every ten seconds gets spotted right away. If gestures really follow the text, the formats where you watch someone talk for a long time benefit: tutorials, training, product walkthroughs, translated company videos. It is also what makes a 10-second video ad believable when someone talks to camera.

For film or fiction, though, Avatar 6 is still a presenter tool: one framed person talking, not a scene. To stage characters, you still need to pick your video model shot by shot.

How to read the 84.4% figure. Imaginode diagram.Source: heygen.com

What I would do this week

The first video is free: use it for a real test, not a demo.

**Take care of the source recording.** Good light from the front, a plain background, your real voice, a few natural gestures. Everything the model learns comes from there.

**Keep your own voice.** HeyGen's own table shows it: with the person's voice, the avatar passes far better than with a synthetic one.

**Write a script that calls for gestures.** A three-point list, a "on the left, on the right" comparison, a question to the viewer. That is where the new feature shows, or does not.

**Show the result to someone who does not know.** It is the only test that counts, and it is the one HeyGen ran: just ask them whether the video is real.

What we don't know

The page describes the model, not yet its limits.

No independent test has confirmed HeyGen's scores yet, nor the "#1 on Avatar Arena" ranking quoted in the press release. We also don't know how many credits a minute of Avatar 6 uses, or how it handles languages other than English, since the page only shows English examples. To make a photo talk directly on the canvas, without building an avatar, Imaginode offers HeyGen Video 1.

How we analysed this

Written on November 18, 2026 from HeyGen's Avatar 6 page and the press release published on PR Newswire on November 17, 2026, both read on November 18. The screenshots come from those two pages, taken the same day.

We have not tested Avatar 6. The scores quoted are the ones HeyGen publishes from its own tests: no independent body has checked them yet. Disclosure: Imaginode offers HeyGen Video 1, another model from the same company, on its canvas, but not Avatar 6.

Want to try this on a real canvas?

Everything described in this article runs on Imaginode, in your browser.

Get started