A Guth Labs publication ↗
Back to News

HeyGen launches Avatar 6 with script-aware gestures and two-person scenes

Two empty cream and sage podiums connected by glass sound waves and a gold arc.
AI-generated editorial illustration

HeyGen launched Avatar 6 on November 17, adding gestures that respond to a script and support for two avatars speaking in the same scene. Its product page describes controls for emotion, expression and movement, alongside API and MCP access.

The company's technical report says the model generates frames at 2880×1620 and can sustain a continuous performance for the length of an audio track, without a fixed duration limit. A photo can supply the visual reference, with a recording optionally providing more identity information.

HeyGen's headline realism score needs context. Its reported 84.4% figure is twice the share of answers in which raters selected the generated clip as real in paired comparisons. The underlying result was 42.2% across 270 answers, not a finding that 84.4% of viewers mistook every avatar for a person. These are company-reported tests.

For avatars made from someone's footage, HeyGen says it requires a live consent recording with a session-specific code and an identity match before rendering. The release matters for training and presentation workflows that need controllable delivery, while its evaluation details make the realism claim easier to interpret.

Sources and citations

Revision history

Published .