Google Gemini Generates 2 Avatar Videos for $20 Subscribers, Testing Omni Deepfake Guardrails
Updated
Updated · WIRED · May 21
Google Gemini Generates 2 Avatar Videos for $20 Subscribers, Testing Omni Deepfake Guardrails
4 articles · Updated · WIRED · May 21
Two 10-second Gemini clips turned Reece into a digital avatar—singing to a dinosaur in Dolores Park and surfing under the Golden Gate Bridge—showing Google’s new Omni video model can produce a close likeness with minimal setup.
A roughly 5-minute enrollment had him read two-digit numbers and turn his head for the camera, but the $20-a-month AI Pro plan hit usage caps after just two videos, with limits resetting every 5 hours.
The outputs mixed photorealistic backgrounds and convincing facial resemblance with glitches: stuttered speech, odd teeth, random props and nonsensical clothing, including a denim surfing outfit.
Google says only adult users can generate videos with their own avatar, a tighter rule than OpenAI previously used, as generative-video tools face growing scrutiny over nonconsensual deepfakes targeting women.
Despite the errors, the test suggests consumer AI clones are becoming more lifelike and easier to make, sharpening both the appeal and the unease around personal digital doubles.
As AI avatars industrialize fraud, can Google’s safety measures prevent the weaponization of synthetic trust?
If AI avatars are still 'cringeworthy,' is the hype for a seamless digital reality outpacing the flawed technology?
Beyond the $20 fee, what are the hidden environmental and data privacy costs of creating your digital twin?