HeyGen’s Avatar 6 brings 2K video and selective regeneration
HeyGen’s Avatar 6 changed 34.4% of frames in selective edits, generated native 2K video with script-timed gestures, and won New Measure’s early preference tests.
By Amara Osei
4 min read

What's News
- HeyGen introduced Avatar 6 in a November 17, 2026 technical report, offering native 2K output and no fixed duration limit.
- Selective edits changed 34.4% of frames on average, versus 99.7% to 100% for four rivals.
- HeyGen reported an 84.4% indistinguishability score from 270 side-by-side answers.
- New Measure ranked Avatar 6 first with Elo scores of 1,197 and 1,237.
- HeyGen personal-avatar videos typically took three minutes, versus one minute for Runway GWM-1 Avatars.
By Amara Osei
HeyGen said Avatar 6 changed an average of 34.4% of frames in a 20-edit comparison, against 99.7% to 100% for MiniMax H3-max, Wan 3.0, Seedance 2.5 and HappyHorse Edit. HeyGen says the narrower regeneration preserves the surrounding take while correcting words or adding gestures.
What did HeyGen release?
HeyGen introduced Avatar 6 in a technical report published November 17, 2026. The audio-driven model generates a specific person from one photograph, with an optional short identity recording, and produces native 2K output without a fixed duration limit.
HeyGen says Avatar 6 reproduces a person’s appearance, manner of speaking and habits of motion together. It became available in HeyGen’s web app, mobile app and API in mid-November 2026. Its system combines a generator, a super-resolution stage that restores details such as skin texture and hair, a motion planner and HeyGen Voice for speech in the speaker’s own voice.
HeyGen offers Avatar 6 Quality or Avatar 6 Turbo. The company says Turbo uses its accelerated inference stack for lower latency.
How do the gestures work?
According to HeyGen’s product page, Avatar 6 uses a script’s meaning to place pointing, counting or directional movement. Users can separately specify mood, facial expression and hand movement, or attach an action to a phrase through written prompts. Examples include waving during a greeting and counting on fingers through a list.
HeyGen says its motion planner times those actions to the relevant words while preserving the subject’s characteristic movement between them. The company’s demonstrations include continuous 25- to 28-minute videos generated from one audio track without loops, so runtime follows the speech.
HeyGen also showed two speakers in one frame. Its two-minute host-and-guest podcast alternates wide shots with closer views of each person.
How much of the video changes?
HeyGen’s recorded-video editor can replace words or add a gesture, then regenerate only the edited span with newly synthesized speech in the person’s voice. The company called rival approaches “near-total rewrites,” compared with Avatar 6’s 34.4% average frame change in its 20-edit test.
HeyGen reported an 84.4% indistinguishability score. External raters chose Avatar 6 as the real recording in 42.2% of 270 side-by-side answers; HeyGen doubled that rate to 84.4% because a 50% selection rate would equal chance.
The side-by-side test involved 20 external raters and 76 people. In separate single-clip tests, HeyGen reported that 65.1% of answers judged Avatar 6 clips using the person’s own voice as real, versus 78.9% for genuine recordings.
HeyGen’s external-rater comparisons covered only answers with a preference. The company recorded:
- A 50.1% preference for Avatar 6 against Wan 3.0 from a photograph, which HeyGen described as “a tie.”
- A 54.9% preference against Kling 4.0 from footage.
- A 72.4% preference against Seedance 2.5 from footage.
What did independent testing establish?
New Measure’s early Avatar Arena results ranked Avatar 6 first for personal avatars and stock presenters. The model posted Elo-scale preference scores of 1,197 and 1,237, respectively, against an average of 1,000 in each category.
Under New Measure’s protocol, creators compared two videos reading the same script, with provider names hidden and video order randomized. New Measure counted votes and edits from October 27 through November 16, 2026.
New Measure cautioned that participants opted into the arena and that the typical script contained 62 words. Both videos lasted 35 seconds or less in 97% of completed comparisons, leaving long videos and languages with insufficient testing unproven.
Preference did not translate into faster delivery. New Measure found that HeyGen personal-avatar videos typically took about three minutes, versus about one minute for Runway GWM-1 Avatars; one in 10 exceeded nine and a half minutes. New Measure excluded avatar setup and speech preparation from those timings.
What do the APIs and consent rules allow?
HeyGen’s API documentation lists heygen-avatar-6 and heygen-avatar-6-turbo, with 1080p as the default resolution. A Model API request combines a reference image or video with speech audio; an Application API request starts with a HeyGen avatar look and a script with voice or audio.
HeyGen’s Application API supports Avatar 6 only for single-avatar requests. The company excludes raw-image jobs, studio scenes, batch creation and scene editing; when customers omit engine selection, the application continues to use Avatar IV.
For footage-based avatars, HeyGen requires the person to record an on-camera consent statement with a session-specific one-time code before rendering. HeyGen verifies that the consent speaker matches the training-footage subject.
Under HeyGen’s access rules, photo references are available to all API workspaces with the person’s permission, while realistic video references remain enterprise-only. HeyGen’s documentation says video-reference consent is verified before submission, although some contracts can proceed without that step after case-by-case review.
For production teams, the next test is long-form multilingual performance: New Measure favored Avatar 6 on short clips but did not establish long-video or language results. Its roughly three-minute typical generation time also leaves speed behind Runway’s one-minute benchmark.
Source: unite.ai
More from Amara Osei
Show full bio
Senior reporter covering companies, deals and strategy in entrepreneurship, startups and venture funding at Business Bearings.
2,248 articles