HeyGen Introduces Avatar 6, an Audio-Driven, Context-Aware AI Model

Article Featured Image

HeyGen, a personal avatar company, has launched Avatar 6, an artificial intelligence avatar model to deliver realistic human movement in AI-generated video. The new audio-driven model interprets user scripts and generates corresponding gestures.

Avatar 6 generates movement based on the context and delivery of speech, allowing gestures to mirror how a person would actually communicate. HeyGen uses footage of each individual to understand how they naturally look, move, and express themselves. Users can create an avatar with as little as 30 seconds of video, with the option to upload additional footage manually or from social media to create the most realistic, varied movements possible. Users can connect their personal YouTube, Instagram or TikTok accounts to create avatars from existing footage.

A tenfold increase in the model's runtime context window, from approximately three to 30 seconds, enables more realistic, varied movements based on each user's actual footage, with improved lip synchronization and native 2K video output.

"People decide in seconds whether a video feels like a person talking to them, and when it doesn't, they stop listening," said Joshua Xu, co-founder and CEO of HeyGen, in a statement. "For professionals whose businesses depend on trust, a video that feels 'off' can do more harm than good. We started HeyGen because we believed people should be able to share their ideas and expertise without having to constantly get in front of a camera. Avatar 6 helps them create high-quality content that genuinely reflects who they are."

With Avatar 6, third-party platforms can generate photo-based and virtual avatars, while realistic videos of real people remain within HeyGen's platform, where explicit consent is required.

"Modeling human realism remains one of the hardest challenges in AI video," said Rong Yan, chief technology officer of HeyGen, in a statement. "With Avatar 6, we've made significant progress in how accurately we can model human looks and movement. It also is able to capture the complexity of human expression while giving people greater control over how it's represented on screen. In our internal Turing testing, 84 percent of participants rated the AI avatar as equally realistic as the actual person. Testing also revealed that the new model outperforms competing video and avatar models in the market."