Introducing Gemini Omni: A Versatile New Family of AI Models Designed to Generate Anything

Google has unveiled an exciting new lineup of generative AI models known as Gemini Omni. The first model in this series, Omni Flash, is capable of creating AI-generated videos from a variety of inputs, including text, images, videos, and audio. The vision for Omni is ambitious, aiming to “create anything from any input,” which is reflected in its name.

Omni Flash is being positioned as a video counterpart to the Nano Banana image generation model, which has already produced over 50 billion images since its launch last year. With Omni Flash, users will have the ability to insert their likeness into videos, a feature that may not appeal to everyone, but has seen popularity with the Nano Banana model, according to Nicole Brichtova, who leads the product team for Omni.

With Gemini Omni Flash, users can generate clips that are up to 10 seconds long, combining both video and audio, as shared by Dumitru Erhan, senior research director at Google DeepMind. The team is also working on extending this duration in the future.

While Google already has a video generation model called Veo, which focuses on text-to-video creation, Omni Flash takes it a step further by allowing users to use existing videos as a foundation for new creations. Additionally, Omni Flash boasts significantly more world knowledge than Veo, thanks to the extensive training data from Gemini, as noted by Koray Kavukcuoglu, CTO of Google DeepMind.

Gemini Omni Flash will be accessible starting Tuesday through the Gemini app, Google Flow, and YouTube Shorts.