Google's New 'Gemini Omni Flash' Model Can Generate "Anything From Any Input"

In addition to Gemini 3.5 Flash, Google has also unveiled Gemini Omni at Google I/O 2026, and this might be one of the company’s wildest Gemini announcements yet. Unlike previous Gemini upgrades that mainly focused on chat, coding, or reasoning, Gemini Omni is built around creation. Google says the new model can generate “anything from any input,” starting with video.

The first release under this new family is called Gemini Omni Flash, and it’s rolling out today through the Gemini app, Google Flow, and even YouTube Shorts. According to Google, the model can combine text, images, audio, and video together to generate fully edited video clips that stay visually consistent across multiple prompts and edits.

Gemini Omni turns video editing into a conversation

Gemini Omni Flash reportedly lets users modify videos simply by talking to it naturally. That means users can upload footage and ask Gemini to change objects, environments, motion, physics, or even entire scenes through conversation. Google showcased examples where statues transformed into bubbles, mirrors behaved like liquid metal, and skateboards gained animated motion effects without rebuilding scenes from scratch.

Interestingly, Google says the model also understands physics and real-world behavior much more accurately than previous systems. The company demonstrated chain reaction marble tracks, claymation-style explainers about protein folding, and rapid-fire alphabet visualizations generated from single prompts.

Speaking of prompts, Gemini Omni can reportedly pull from almost any reference type. Users can combine images, videos, music, drawings, and written instructions together into one cohesive clip. Google says voice references for audio are supported first, while broader audio input support is coming later.

Google is already pushing Omni into YouTube

Google has also announced digital avatars powered by Gemini Omni. Users can create AI-generated video versions of themselves using their own voice and likeness, although the company says broader speech-editing capabilities are still being tested carefully. Every video generated through Gemini Omni will include Google’s SynthID watermarking system as well.

Gemini Omni Flash is rolling out globally today for Google AI Plus, Pro, and Ultra subscribers, while YouTube Shorts and the YouTube Create app are getting access free of charge starting this week.