Synthetic avatar deepfake

A deepfake that generates an entirely fabricated person, including movement and expression, rather than swapping a face onto real footage.

No real video sits underneath it. The software creates the face, the body, and the motion from scratch, and it can respond to a conversation as it happens.

A deepfake subtype that renders a fully synthetic human, body and face, in real time, rather than modifying existing video, making it viable for live impersonation on a video call rather than only pre-recorded content.

A synthetic avatar attack typically works like this:

1. The attacker trains or configures a model on images, video, and audio of the person to imitate.
2. The software renders a synthetic person that moves and speaks.
3. The attacker feeds the avatar into a video call, often through a virtual camera.
4. The avatar takes part in the call as if it were the real person.

The approach differs from other deepfake types in a few ways:

• No source footage, since the avatar is rendered instead of edited from a real video.
• Live operation, since it can respond in real time instead of playing a prepared clip.
• Combined channels, since it can pair a synthetic face with a cloned voice.

Synthetic avatar deepfakes sit next to terms that are easy to mix up:

• Face-swap deepfake modifies real footage, while an avatar is generated from scratch.
• Deepfake is the broader category.
• GAN (Generative Adversarial Network) is one of the techniques that can generate synthetic faces.

For more, see What Last Year’s AI Deepfake Fraud Cases Can Teach Us In 2026.