Google launches Gemini Omni for video generation

Google unveiled Gemini Omni at its I/O 2026 developer conference as a multimodal model that can generate and edit video and other media from a wide range of inputs. The first release, called Gemini Omni Flash, will be available inside Google Flow and Flow Music for Google AI subscribers.

Onstage, DeepMind CEO Demis Hassabis described Omni as “our new model that can create anything from any input.” Google says the system merges Gemini’s reasoning and world knowledge with its generative media tools to enable more advanced multimodal creation and editing.

Gemini Omni integrates several of Google’s media models: Veo for video generation, Nano Banana for image editing and Genie for broader generative tasks. Google describes the integration as a way to extend image-editing capabilities into video and to keep characters, backgrounds and movement consistent across edits.

Demonstrations at I/O showed Omni producing a claymation-style educational video explaining protein folding and performing conversational edits on a selfie video, where visual elements and the environment were changed through natural-language prompts. Google says Gemini’s reasoning lets Omni follow broad instructions without requiring users to specify every low-level detail.

Google also introduced Flow Agent, an assistant inside Flow that can brainstorm scenes, organize creative assets, recommend plot changes and apply batch edits. Flow Tools will let users create custom editing workflows using natural-language prompts without writing code. Google said Omni Flash will appear in Flow and Flow Music first, with broader access and features to be added over time.

Nano Banana was previously used for meme generation and conversational image edits and helped increase adoption of Gemini last year. Google plans to bring many of Nano Banana’s editing features into video through Omni.

Google framed Omni as an early phase of a larger effort to build a model that can understand and simulate complex scenes, starting with video and expanding capabilities and access over time. Hassabis noted that building Gemini as a multimodal system from the start was intended to support that direction.