MOUNTAIN VIEW, CALIFORNIA — Google introduced an avatar feature for its Flow video tool at the company's annual I/O developer conference in Mountain View, California. Elias Roman, vice president of product management at Google Labs, demonstrated the feature using a digital clone of himself created from an earlier likeness scan.

Flow, launched last year under Google's experimental Labs division, lets users generate and remix AI videos and images. The avatar capability allows users to insert themselves into AI-generated video clips using Google's new Omni Flash model, which succeeds Veo. Google describes these social-first, selfie deepfakes as avatars rather than cameos or characters.

"This is for creators who want to bring themselves into their content but don't want to have to shoot themselves," Roman said during the demonstration. He added that "Google has never had a product line for creative work before."

"You can capture your voice and your visual identity from multiple angles and have that show up with pretty high levels of fidelity," Roman said.

To set up an avatar, users open Flow account settings and scan a QR code on their phone. Google then asks them to record themselves saying a string of numbers aloud and to move their head to capture every angle. The selfie-capture style is similar to the one used by OpenAI's Sora app, an AI-first social media platform where people could generate and share clips of themselves. OpenAI launched the Sora app last year and wound it down after less than seven months. Avatars are available through the Gemini app and YouTube.

Google's initial focus with avatars is to let users create AI versions of themselves only and not other people. Every video generated with the Omni model includes Google's SynthID watermark.

Roman generated a tongue-in-cheek clip of himself teasing the Flow team in front of a dumpster fire, then used Flow to request changes including a new background and a different-colored shirt. Omni Flash adjusted the clips while preserving the avatar's details. According to Roman, the model is a major improvement over earlier versions, in which created characters could warp during successive video generations.

The Omni Flash model overhauls video generation with richer detail throughout clips, and Flow users can generate characters with more consistency. Users can also repeat custom instructions when generating videos and create automated workflows that sort similarly styled clips into folders.

The changes are part of Google's effort to make AI agents, which it defines as automated software taskmasters, and vibe coding, which it defines as building bespoke features with natural language prompts to AI, more mainstream for a broader audience. Last month, YouTube Shorts added a limited option for users to make similar AI avatars for insertion into clips. Last year, Meta rolled out an AI feature that can translate Instagram Reels into different languages and adjust creators' lips to match the different voices.