Gemini Omni Introduction Shows Multimodal Video Creation

A short Google DeepMind introduction released for Google I/O 2026, presenting Gemini Omni’s approach to video creation and editing. It highlights a model designed to use text, images, audio, and video

A short video unveiled by Google DeepMind for Google I/O 2026 introduces Gemini Omni as a video-focused multimodal generation and editing model. Rather than being limited to creating scenes from text alone, the model aims to take images, audio, and existing video as inputs as well, combining them into a single result.

The applications presented in the official introduction focus on conversational editing. Users can change the look, motion, or effects of existing video through natural language and maintain the context and elements of the same scene while continuing to edit. It also includes replacing people or objects with different reference images and creating new videos based on the movement in a drawing.

For video production, Omni aims to use real-world motion such as gravity, movement, and fluids, along with historical, scientific, and cultural context. Google DeepMind explains that this will support video creation and editing that reflects the relationships among the input materials, going beyond realistic rendering. Generated or edited results receive SynthID digital watermarks and C2PA Content Credentials.

Maru