Google Launches Gemini Omni for Video Generation and Conversational Editing
Google used its I/O 2026 conference to introduce Gemini Omni Flash, a multimodal model aimed squarely at video creation and editing. The model takes any combination of text, images, audio, and existing video clips as input, then generates new video or modifies existing footage through a conversational interface - users describe what they want changed, and the model responds accordingly, iterating without requiring new prompts from scratch each time.
The name reflects a broader ambition: Google describes Omni as a framework for generating "anything from any input," with video as the initial focus. The company says the model has stronger grounding in real-world knowledge - history, physics, cultural context - compared to prior video models, which it expects to translate into more plausible motion and scene dynamics. Whether that holds up under broader use remains to be seen, but the claim speaks to a known weakness in current video generation tools.
On the distribution side, Gemini Omni Flash is launching first to Google AI Ultra subscribers globally, with a consumer rollout expected to follow. All output automatically carries a SynthID watermark, tying it into Google's expanding provenance infrastructure. The model also supports avatar-based video creation, letting users synthesize a digital likeness rather than appearing on camera directly.
Google's timing is deliberate - Omni arrives alongside several other I/O announcements including the Google Pics image editor and Genie 3 world model expansion, signaling a coordinated push across the generative media stack rather than a single standalone release.
