Google's Gemini app now lets subscribers create AI video avatars of themselves, powered by the Omni video model.

Google has introduced an avatar feature in its Gemini app that generates AI videos of users' digital clones. The feature is powered by Gemini's new Omni video model and is available only to subscribers of Google's AI Pro plan ($20 a month).

Setup takes about five minutes through the Gemini app: users sit in a well-lit room, point their phone's camera at their face, read a string of two-digit numbers, and slowly turn their head left and right. Unlike OpenAI's earlier approach, Google only permits adult users to generate videos using their own avatar, not others' likenesses.

The generated videos blend photorealistic backgrounds with the user's avatar, though the clips contain errors and inconsistencies—stuttering speech, nonsensical clothing, and jumbled moments. Google states it tries to prevent harm while avoiding blocking benign uses, according to Nicole Brichtova, who leads the Omni product team at Google DeepMind.