Gemini Omni Flash 1.1 is a multimodal model from Google designed to generate, edit, and extend video content using a variety of inputs, including text, images, and existing video clips. Unlike many contemporary video models that prioritize single-shot generation—where a single prompt produces a finished clip—this model is built for a conversational workflow. This means users can iterate on a scene over multiple turns, with each subsequent edit building upon the previous one while maintaining consistency in characters, objects, and environments. The model also features native synchronized audio, meaning that sound and on-screen text are generated as part of the same process rather than requiring a separate post-production pass.
The model's primary differentiator is its grounding in Gemini's broader real-world knowledge base, which includes an understanding of physics, history, science, and culture. This is intended to produce more believable scenes and regionally accurate settings, such as local signage and native languages. Technically, the model supports several specific production workflows: 'first and last frame' control, where a user pins the beginning and end of a shot; video extension, which can push a clip past its initial 10-second limit up to a total of 30 seconds; and character or object swaps. In these swaps, the model can replace a subject while carrying over the original motion and dialogue seamlessly.
To balance high-quality output with production costs, Gemini Omni Flash 1.1 introduces a tiered resolution system. Users can iterate on 'takes' at a 360p draft tier to refine the scene at a lower cost before rendering the final keeper at 1080p or 4K. While the model is positioned as a versatile production tool, the source material does not provide independent benchmark scores to verify its specific performance relative to competitors like Kling or Veo. It relies on the integration of Gemini's existing knowledge to provide its 'grounded' edge, focusing on the continuity of a multi-turn editing experience rather than just the quality of an isolated clip.