HeyGen claims 84% of viewers mistook its Avatar 6 AI avatar for a real person
HeyGen has released Avatar 6, a model that turns a script into a video with a digital presenter. The company claims 84% of viewers could not tell the avatar from a real person, and that the Avatar Arena ranking put the model in first place. An expert needs only a photo and a clip of up to 30 seconds to produce instructional videos without filming. HeyGen itself supplied the 84% figure, so test the quality on the free plan first.
AI-processed from TG: @neyr0graph (NeuroGraph); edited by Hamidun News
Imagine you are an expert or a product manager who has to record training videos every month. You need lighting, a microphone, retakes and editing, and after any change to the text you have to shoot everything again. HeyGen offers a different route: you provide a photo and a short video once, then write a script, and your digital copy "reads" it to the camera.
The new Avatar 6 model is the latest attempt to make such a presenter indistinguishable from a real person. The headline number of the announcement: according to HeyGen, 84% of viewers did not realize they were watching an AI avatar rather than a human. This matters to you if you make courses, instructions or presentations and want to save on filming.
But so far only the company itself has reported the figure, so it needs to be checked.
What HeyGen claims
The company has introduced Avatar 6, a model for videos with digital presenters. Besides the 84% of viewers who could not tell the avatar from a human, HeyGen says the model took first place in Avatar Arena by New Measure. This is a ranking in which avatar models are compared against each other. The source does not specify how the viewer survey was conducted: how many participants there were or how long the videos were.
There are three updates.
First, realism. According to the company, the avatar looks so natural that it can be mistaken for a real person.
Second, gestures that match the meaning. The model adjusts movements to the content of the script. For talking-head video this is a significant detail: if a hand goes up at the wrong moment, or the presenter keeps repeating the same sweep of the arm, viewers quickly sense something is fake. When a gesture lands on the right phrase, the speech sounds more convincing.
Third, a short source clip. As a reference, the company specifies a video of up to 30 seconds and a photograph.
How to use it and what it costs
A typical workflow looks like this. An expert records a short clip and sends a photo. Then they write the text: instructions for customers, a module of an online course, a review of a new release or a presentation for the team. The model turns the script into a video with a digital copy of the author. If the text contains a mistake, you only need to fix the script and generate the video again.
The most interesting question here is how the model handles pauses, facial expressions and gestures in long speech. Details like these determine whether viewers will want to keep watching the avatar beyond the first few seconds.
HeyGen plans with monthly billing:
- Free — $0: 3 videos a month, up to a minute each, with a watermark.
- Creator — $29 (about 2 391 ₽ a month): 600 credits, Avatar 6 Turbo, 1080p export. With annual billing it is $24 (about 1 979 ₽) a month, but the full year is charged up front.
- Pro — from $49 (about 4 040 ₽ a month): from 1 000 credits, Avatar 6 Pro, 4K export. Packages go up to $4 300 (about 354 547 ₽) a month for 100 000 credits.
- Business — $149 (about 12 285 ₽ a month): 1 500 credits and team features. An additional seat costs $20 (about 1 649 ₽) a month.
Note that model versions are tied to plans. Turbo is included in Creator, while Avatar 6 Pro comes with Pro. If you need maximum quality, the cheapest plan may not be enough.
What to check yourself
The 84% figure and the first place in Avatar Arena come from HeyGen's announcement, which includes no independent verification. Treat them as a marketing claim. Such tests usually evaluate short clips, while a video several minutes long may give itself away through repetitive facial expressions, odd pauses or intonation that does not match the meaning.
A practical approach is to take the free plan, upload your own photo and a short video, and generate the same script at different lengths. Show the result to a few colleagues or clients without telling them it is an avatar, and see at which minute they begin to doubt. Think through the ethics separately: for training materials and customer videos, many companies prefer to say openly that the presenter is digital. Brand trust is easier to lose than the money saved on filming.
Bottom line: Avatar 6 needs only a photo and 30 seconds of video to get started and promises more meaningful gestures. Whether that is enough to keep a video from putting viewers off will be shown only by a trial on your own content.
- Nov 20, 2026Suno adds Visuals: turn a song into a clip of up to 15 seconds with lip-sync, for separate credits
- Nov 20, 2026Gemini Decision 1 picks answers instead of writing them: $0.10 per million input tokens, output is free
- Nov 20, 2026A Seven-Minute AI Film in 7 Days for $300–400: Blender Blockouts Instead of Live Shooting
- Nov 20, 2026Suno adds lip-synced music clips of up to 15 seconds to its AI songs
Key Questions Answered
What HeyGen claims?
The company has introduced Avatar 6, a model for videos with digital presenters. Besides the 84% of viewers who could not tell the avatar from a human, HeyGen says the model took first place in Avatar Arena by New Measure. This is a ranking in which avatar models are compared against each other. The source does not specify how the viewer survey was conducted: how many participants there were or how long the videos were. There are three updates. First, realism.
How to use it and what it costs?
A typical workflow looks like this. An expert records a short clip and sends a photo. Then they write the text: instructions for customers, a module of an online course, a review of a new release or a presentation for the team. The model turns the script into a video with a digital copy of the author. If the text contains a mistake, you only need to fix the script and generate the video again. The most interesting question here is how the model handles pauses, facial expressions and gestures in long speech.
What to check yourself?
The 84% figure and the first place in Avatar Arena come from HeyGen's announcement, which includes no independent verification. Treat them as a marketing claim. Such tests usually evaluate short clips, while a video several minutes long may give itself away through repetitive facial expressions, odd pauses or intonation that does not match the meaning. A practical approach is to take the free plan, upload your own photo and a short video, and generate the same script at different lengths.
Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews
Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.