Gemini Omni Flash 1.1

Gemini Omni Flash 1.1 is the second release of Google's multimodal video model, shipped in late August 2026. It generates video with synchronized native audio from text, images, and video references, and it edits and extends footage through plain-language instructions. Version 1.1 adds a resolution ladder up to 4K, working video references, first-to-last-frame interpolation, and extensions that take a clip to 40 seconds.

Four things. Resolution is now selectable from 360p to 4K where the first release was fixed at 720p. Video extension is live, growing a clip ten seconds at a time to 40 seconds, where the original capped at a single ten-second generation. Reference videos now actually steer the output, up to three clips of three seconds each. And image-to-video accepts an end frame, so the model can interpolate between two stills you provide.

Four options: 360p, 720p, 1080p, and 4K, in 16:9 or 9:16. The model renders natively at 360p and 720p, and produces 1080p and 4K as upscales of that generation. In practice that means drafting cheap at 360p, locking the take at 720p, then requesting the delivery resolution once the shot is right.

A single generation runs 3 to 10 seconds, with 8 seconds as the default. From there, extension grows the clip ten seconds at a time up to 40 seconds total, keeping characters, motion, and audio continuous across each join. Extension appends to the end of the clip only; it cannot prepend footage or lengthen the middle.

You ask for a continuation in plain language, like "extend this video" or "the scene continues as the camera pans across the mountains". The model reads the last ten seconds of the clip as context and generates the next beat, lightly adjusting the final frames so the join is seamless. It works on clips the model generated and on footage you upload, as long as an uploaded clip is ten seconds or shorter.

Up to ten reference images and up to three reference videos of three seconds each, alongside the text prompt. Images can pin a character, a product, a setting, or a style, and tags in the prompt assign each one its role. Video references carry a likeness or a motion; any audio inside them is ignored, and asking the model to reason across several full videos at once is not supported.

Each follow-up prompt edits the previous result instead of generating from scratch, and the model keeps the scene state between turns. Short instructions work best: "make the phone invisible, keep everything else the same" beats a paragraph describing the whole frame. Because every turn builds on the last, one focused change per turn keeps characters and motion stable across a chain of edits.

Yes, sound is generated in the same pass as the picture: dialogue with lip sync, effects, ambience, and music timed to the action. Write the audio you want into the prompt, because when a prompt says nothing about sound the model invents its own track, and a spoken line is as simple as writing what the character says.

Yes. Pick Gemini Omni in the video model picker, write the shot with the sound you want, and generate; the take lands on the Canvas next to results from Veo, Seedance, and Kling for side-by-side comparison. Follow-up prompts edit the same scene, and reference images can ride along with the brief.

No, and the confusion is understandable: both come from Google DeepMind and Omni is often written about as the successor to the Veo line. They are separate models. Veo is the cinematic video family, currently Veo 3.1; Gemini Omni is the multimodal family whose first video model is Omni Flash, now at version 1.1. Google has not released a model called Veo 4, so any "Veo 4" reference you see is either speculation or an informal nickname for Omni.

Veo 3.1 is tuned for single-shot cinematic fidelity; Omni Flash 1.1 is tuned for iteration. Omni is the one you direct: it edits its own output turn by turn, takes mixed image and video references, extends a take to 40 seconds, and holds a character and voice through the process. For a one-off hero shot Veo still leads on polish; for a scene you refine and grow, Omni Flash is the workflow.