Gemini lets friends share avatars to appear together in AI videos

Google has opened avatar sharing in Gemini, allowing friends to include one another’s digital likenesses in AI-generated images and videos. People can share a personal avatar and tag each other in prompts to appear together. In its October 29 announcement, Google said the feature maintains character and voice consistency across creations.

Personal avatars require a personal Google Account and a user aged 18 or older. They currently support English and are unavailable in the European Economic Area, Switzerland and the United Kingdom. Making videos with an avatar also requires a Google AI plan.

Owners address invitations to a recipient’s Gmail account. Once accepted, a shared avatar becomes available for that recipient’s image and video requests. The owner can inspect who has access, remove a person and change an individual recipient’s expiry date.

The input recording captures the user’s own face and voice. Google’s selfie-video guidance requires visible eyes, nose and mouth, with no other people or pictures of faces in the background. Recording also requires camera permission and an internet connection. Sunglasses, masks and hats must be removed. This short personal recording supplies the reference used to create the avatar.

Google’s Nano Banana documentation says an avatar can be reused for personalized images without uploading a fresh photograph every time. Image creation also supports targeted edits and combining multiple reference images. Shared avatars extend that workflow by allowing another person’s permitted likeness to appear in the resulting composition. Image output has separate plan limits: Google documents 1K downloads without an AI plan and 2K with one. Avatar-sharing permission and download resolution follow their respective availability conditions.

For video, Gemini Omni accepts text, images and video references, and supports follow-up edits within a conversation. Users can change characters, objects or camera angles and ask for generated audio. The computer help page permits one video and up to five reference images per request. Uploaded images require the relevant usage rights, and Google’s content policies apply to generation. Text-only requests default to landscape; reference media determines the aspect ratio when supplied.

Ending access leaves existing creations in place

Revoking permission prevents the recipient from making further content with the avatar. Google’s privacy guidance says images and videos already created remain available. It also says the owner has no control over what a recipient generates with a shared likeness.

Recipients cannot access, edit or delete the original likeness recording. That recording is stored in its owner’s Google Account. Google says recordings are automatically deleted after three years without access or use.

Flow sets its own boundary around distribution. Its help page allows a personal avatar to serve as a reference in projects, but prohibits sharing avatar data through Flow Tools or public project links. Generated likeness content undergoes safety checks. Avatar setup can begin on a computer but requires a mobile device to finish.

Google uses avatar recordings for content generation and safety. Human reviewers may help check that a real person has authorized creation of their avatar. Google excludes these recordings from generative-model training except to address abuse or harm. Some derived face and voice data may qualify as biometric information under local laws.

Deleting a recording stops new content creation with that avatar; using it again requires a new face and voice recording. Existing posts require deletion in their host service. The deleted face and voice recording is removed from Google’s systems.