Introducing Gemini Omni

Details

Gemini Omni is DeepMind's new model family that brings Gemini's reasoning capabilities to multimodal generation, starting with video. The first release, Gemini Omni Flash, allows users to combine any mixture of text, images, audio, and video as input, then generate or edit high-quality video through conversational natural language instructions.\n\nKey capabilities include multi-turn edit consistency — characters, physics, and scene context persist across edit sessions — targeted transformation (change specific objects or entire environments), and a reimagining mode that turns existing footage into scenes that could never have been filmed. Every instruction builds on the last, without losing the thread of the original scene. Future Omni releases will extend to image and audio output modalities, completing the full generative stack.\n\nThe release follows Nano Banana (image generation and editing) and represents the next layer of Gemini's natively multimodal architecture. Omni Flash is rolling out in the Gemini app, Google Flow, and YouTube Shorts today — putting conversational video generation in front of hundreds of millions of short-form creators and Gemini users simultaneously. This is Google's direct competitive response to dedicated video generation models.