Google is giving its AI filmmaking tool another major push forward. At its I/O developer conference earlier this year, the company introduced Gemini Omni Flash inside Google Flow, its browser and mobile app for generating and editing video with AI. Now Google is rolling out an update to that model, called Gemini Omni 1.1 Flash, that adds a set of controls aimed squarely at people who want their AI generated footage to look less like a demo and more like something that could survive contact with a real production pipeline.
Google Flow started out as something closer to a filmmaking playground built around Veo, Google’s video generation model, giving creators tools for camera movement and simple shot extension. Over time, Google folded image generation and editing into the same workspace, and then layered on Gemini Omni, a model built to combine text, images, audio and video in a single creative session. Omni 1.1 Flash is the next step in that evolution, and it is arguably the update that makes Flow feel like a production tool rather than a sandbox.
Longer scenes that actually hold together
The headline feature is scene extension, and it is not new to Flow. What has changed is how much of the existing footage the model actually looks at before continuing it. Previous versions of the model could only reference the final second of a clip when generating what came next, which made it prone to drifting off model, changing a character’s face, wardrobe or the mood of a scene the moment the camera cut continued. Omni 1.1 Flash can now analyze up to ten seconds of prior footage before it generates a continuation, which gives it far more to work with when it comes to keeping faces, lighting, and narrative tone consistent from one clip to the next.
Extensions can be chained in ten second increments, up to a cumulative total of forty seconds. That is still a far cry from a full length short film, but it is enough room to build a scene with an actual beginning, middle and end, or to branch a story off in a new direction once the first clip is locked in.
Pinning the start and the end of a shot
The other big addition is first and last frame control. Rather than describing a whole shot in a text prompt and hoping the camera move lands the way it was imagined, creators can now specify the exact frame a shot should start on and the exact frame it should end on, and let the model generate everything in between. Google is positioning this as the tool for the kind of shots that are genuinely hard to prompt your way into, like a slow camera orbit around a subject, a zoom that pushes through one scene into another, or a loop that needs to end exactly where it began so it can play seamlessly.
It is a small feature on paper, but it addresses one of the more persistent frustrations with generative video: the gap between what a filmmaker can picture and what a text prompt can actually communicate. Locking down the two endpoints of a shot turns a vague verbal description into something closer to a storyboard.
A cheaper way to iterate before committing to a final render
Generative video has always had an awkward tradeoff built into it. High resolution renders look better, but they also cost more credits and take longer to generate, which makes it expensive to iterate on an idea. Omni 1.1 Flash tries to soften that tradeoff with a new 360p draft mode. Google says these lower resolution previews render up to sixty percent faster and cost about a third as much as a standard 720p generation, based on the model’s system throughput at each resolution.
The idea is that creators can burn through several rough versions of a shot at low cost and low commitment, and only spend the credits on a full quality render once they have found a version worth keeping. Inside the Google Flow mobile app specifically, this workflow is aimed at people who want to sketch out ideas away from a desktop, drafting concepts on a phone and then upscaling their favorite take later. On the output side, Flow users can export finished video in 1080p or 4K, resolutions that Google says are meant to be ready to drop straight into higher end digital, social or broadcast editing workflows without needing an additional upscaling pass somewhere else.
Feeding the model a reference clip
Omni 1.1 Flash also adds support for video references as part of its multimodal input. Creators can hand the model up to three seconds of a separate video clip to use as a reference when generating a new scene, which helps it hold onto visual context and character consistency drawn from that footage. It is a useful complement to scene extension, since it means consistency does not only have to come from continuing an existing Flow generated clip. Footage from somewhere else can now inform a new one.
Built for developers too, not just Flow users
Alongside the consumer facing rollout in Flow, Google is pushing Omni 1.1 Flash out through the Gemini API in Google AI Studio, describing it as production ready for professional use by developers building their own generative video workflows, creative tools or media editing software. Through the API, developers can call scene extension using a previous_interaction_id parameter that references an earlier generated clip, along with a response_format field that controls the output resolution, whether that is a 360p draft or a full 720p, 1080p or 4K render.
Google has also made the model available through its Gemini Enterprise Agent Platform, and several companies are already building on top of it. Adobe has incorporated Omni Flash into Firefly, while Figma’s creative director has talked about the model letting teams move beyond simply generating videos toward actually directing them. Runway’s chief creative officer has described it as another way for the company’s users to move quickly between ideas.
Rolling out now
The update is live globally today for Google AI Plus, Pro and Ultra subscribers inside Google Flow, and scene extension specifically is also available to those same subscribers inside the standalone Gemini app. On the developer side, Omni 1.1 Flash is available now through Google AI Studio and the Gemini Enterprise Agent Platform, with documentation, a cookbook and prompting guides published alongside the release to help developers integrate scene extension, video references and upscaling into their own applications.
Taken together, the update reads less like a single flashy new trick and more like a checklist of the practical friction points that separate a fun AI video demo from a tool people can actually build something finished with. Longer, more consistent scenes. Precise control over how a shot starts and ends. A cheap way to iterate before paying for a full resolution render. Output that is actually ready for a real editing timeline. None of it is flashy in isolation, but together it pushes Google Flow further into territory that professional creators, not just people playing around with prompts, might actually reach for.