Gemini Omni Agent will launch along with Avatars support

A fresh banner for Gemini Omni has surfaced inside the Gemini web build, still hidden from users but already legible in the code. Added only yesterday, the string invites people to create videos with Gemini Omni, try a template, and place themselves into different scenes, drawing from any mix of images, text, and clips, then refining the result through conversation. If something doesn’t look right, users can simply ask the model to adjust it.

The wording lines up with what model IDs spotted earlier this week implied: Gemini Omni will operate as an Agent rather than a standalone video model, capable of combining multiple modalities into a single video output. That framing also explains why direct comparisons with ByteDance’s Seedance 2.0 have been off the mark. Omni isn’t being positioned as a top-tier generator going head-to-head with the current benchmark leader; it’s an orchestration layer where video is one possible output rather than the whole point.

The reference to inserting yourself into scenes ties the launch directly to Likeness, the avatar system previously branded as Characters and now showing up across Gemini web and mobile under the Avatars label. The feature mirrors what Sora calls characters and what Meta has been building for its own creative surfaces. Users will most likely need a phone to capture a short selfie video for the initial scan, which then becomes a persistent identity tied to their account and can be invoked within generative flows. The UI tag for Avatars recently flipped to “new,” a marker Google typically uses right before a feature goes live.

Timing is the most telling part. The Android Show kicks off today with a tight focus on Android and mobile, which is exactly where the Avatars capture flow lives. If Avatars get their public moment on stage, Gemini Omni could ride alongside them as the agent that actually puts those avatars to work, with Google I/O on May 19–20, leaving room for a deeper developer reveal next week.