HeyGen launched Avatar 6 on 17 November 2026, a new AI avatar model that reads a script and generates gestures and expressions to match what is being said. The model outputs native 2K video, works from as little as one photo or 30 seconds of footage, and is available in the HeyGen app and through its API.
Quick Take
- What: HeyGen’s newest avatar model for talking-head videos, built in-house.
- Headline feature: context-aware gestures such as pointing, counting and directional motion, driven by the audio and the script.
- Realism claim: raters picked Avatar 6 as the real person 42.2% of the time in side-by-side tests, which HeyGen reports as 84.4% indistinguishable.
- Output: native 2K frames, with longer runtime context of about 30 seconds instead of about 3.
- Access: HeyGen app and API, models heygen-avatar-6 and heygen-avatar-6-turbo; existing Personal Avatars upgrade automatically.
In plain English
HeyGen makes digital versions of people that can present a video for them. Older avatars tended to repeat the same few hand movements, which made videos feel fake. Avatar 6 listens to what the avatar is saying and moves the way a real speaker would, for example counting on its fingers when the script lists three things. It also needs very little input to get started, and HeyGen says it checks that the person in the video has agreed to it.
What happened
HeyGen announced Avatar 6 in a press release from Los Angeles and published a technical report the same day. The company calls it its “smartest and most powerful AI avatar model yet.” CEO Joshua Xu said: “People decide in seconds whether a video feels like a person talking to them, and when it doesn’t, they stop listening.” HeyGen also says Avatar 6 ranked first in New Measure’s Avatar Arena, a blind head-to-head comparison.
Background
AI video is moving fast on two tracks. General video models such as Seedance 2.5 and Kling generate any scene, while avatar companies focus on one person talking to camera for training, sales and social videos. HeyGen recently released its own general video model and the HeyGen Voice model. The company says it reaches more than 40 million users and passed $200 million in annual recurring revenue in June.
How it works
- Audio-driven motion: the model interprets the context and delivery of the speech and generates matching gestures, instead of replaying a loop.
- Prompted gestures: users can write cues next to the words, such as a thumbs up on a given phrase, and set a motion prompt and expressiveness in the API.
- Inputs: one photo, or footage of the person; users can also connect YouTube, Instagram or TikTok accounts to build avatars from existing videos.
- Length and quality: the technical report shows continuous videos of 25 to 28 minutes generated from one audio track, at 2880 by 1620 pixels.
- Word edits: changing a word in a recording regenerates only the edited part in the person’s own voice.
- Consent: before an avatar built from footage can render, the person records an on-camera consent statement with a one-time code, and HeyGen checks it matches the training footage.
Why it matters
For creators and small businesses, the pitch is fewer reshoots. HeyGen cites its own survey in which 71.6% of more than 1,000 small business owners had recorded a video and then decided not to post it.
For developers, the API exposes two model tiers, 1080p or 2K output and motion controls, so avatar video can be built into other products.
For companies, more realistic avatars raise the stakes on disclosure and consent. HeyGen keeps realistic videos of real people inside its own platform, while third-party platforms can generate photo-based and virtual avatars.
What’s still open
- The realism and preference figures come from HeyGen’s own tests and from raters it recruited; independent benchmarks will matter.
- How viewers and platforms will label avatar videos that are this close to real footage.
- Credit costs for long 2K videos on paid plans, which start at $29 a month for 600 credits.
Sources
- HeyGen, press release “HeyGen Introduces Avatar 6,” 17 November 2026
- HeyGen Research, “Avatar 6: Talking Avatars That Are More Like You,” technical report, 17 November 2026
- HeyGen, Avatar 6 product page and developer documentation
i10x Perspective
Avatar 6 shows where the avatar market is heading: less about looking real in a still frame and more about moving like a real person over a long video. That makes consent checks part of the product, not an add-on. More launches are in our Video & Image section.



