Gemini Omni Flash is here, and it is Google’s biggest swing at video AI yet
Google introduced Gemini Omni at I/O 2026, a new multimodal model family that creates and edits content from any input you throw at it. The full announcement landed in a new blog post from Google DeepMind, and the first model, Gemini Omni Flash, is rolling out today.You can mix text, images, audio, and video as references, and Omni spits out a polished video grounded in Gemini’s actual world knowledge. Image and audio outputs are coming later, but video is the headliner for now.
Video by Google
Conversational video editing is the real shift
The standout feature is the ability to edit a video by just talking to it. Tell Omni “make the sculpture out of bubbles” or “when the person touches the mirror, make it ripple like liquid,” and the model does it while keeping your characters consistent across every turn.
Video by Google
This is the same secret sauce that has made Nano Banana so impressive at preserving likeness in image edits, now applied to motion and time. The scene remembers what came before, the physics hold up, and you can keep iterating without losing the thread.
Better physics and real-world knowledge
Omni does not just render pretty visuals, it reasons about what should happen next. Google says the model has a stronger grip on gravity, kinetic energy, and fluid dynamics, so scenes feel less like AI hallucinations and more like actual footage.It also taps into Gemini’s broader knowledge of history, science, and culture to generate explainer-style videos. A short prompt about protein folding, for example, can yield a full claymation breakdown without you doing the legwork.
Mix any inputs you want
This is where things get genuinely fun. Hand Omni an image of a character, a reference video for camera movement, and an audio clip for the beat, and it fuses them into a single cohesive output.
Want a walk cycle that style-shifts in sync with the music? Done. Want harp sounds tied to specific motions in your shot? Also done. Voice references are supported for audio at launch, with other audio types rolling out soon.
Avatars and SynthID watermarks
Omni also lets you create videos with your own digital avatar using your real voice. Google is being cautious about editing audio and speech on existing footage, but the avatar feature is here at launch.
Every Omni-generated video carries an invisible SynthID watermark, and you can verify any video’s AI origin through the Gemini app, Chrome, or Search. That is a meaningful guardrail given how quickly this technology is moving.
Where you can use it today
Gemini Omni Flash is rolling out today to Google AI Plus, Pro, and Ultra subscribers globally through the Gemini app and Google Flow. It is also rolling out at no cost on YouTube Shorts and the YouTube Create App this week, with API access for developers in the coming weeks.
My honest take
Nano Banana made me a believer because of how well it holds onto identity across edits. Flow, the last time I used it, felt limited and a bit clunky, so I am genuinely excited to see what it can do now that Omni is under the hood.
#Googles #newest #model #lets #edit #video #talking