Gemini Omni 1.1 Flash: Google Makes AI Video Production-Ready With 40-Second Scenes, Keyframe Control, and 4K

Two years ago, generative video was a party trick: a few seconds of shimmering footage that fell apart the moment a character turned their head. On August 27, 2026, Google shipped an update that makes the strongest case yet that AI video has crossed from demo territory into production territory. Gemini Omni 1.1 Flash — the first major version bump to Google’s Omni video model since its launch earlier this year — is a suite of creative controls aimed squarely at developers building video tools, and its feature list reads less like a model refresh and more like a post-production checklist.

What’s new in Omni 1.1

The headline upgrade is scene extension. Previous models could only reference the final second of a video when continuing it, which meant longer sequences drifted: characters changed appearance, lighting shifted, and stories lost coherence after a few cuts. Omni 1.1 analyzes up to 10 seconds of prior context before generating new footage, keeping character identity, motion, and narrative state consistent. Scenes can be extended in 10-second increments up to a total cumulative length of 40 seconds — long enough for a full narrative beat rather than a glorified GIF.

Second is first and last frame interpolation. Developers can now specify the starting and ending frames of a shot, and the model generates continuous video between those keyframes. This is the kind of control that matters enormously for real editing: camera orbits, whip-pans, dolly zooms, and seamless loops all become promptable by anchoring where a shot begins and where it ends. In Google’s demo materials, a close-up of a drummer whip-pans to reveal a saxophonist and a ballet dancer mid-spin — one continuous shot, no jump cuts.

Third is a 4K upscale path. Generate fast and cheap, then upscale selected results to 1080p or native 4K for final delivery. Google is explicitly optimizing for a draft-then-final workflow: a 360p draft mode generates previews up to 60% faster (by system throughput) and at one-third the cost of 720p. For teams used to burning budget on full-resolution generations just to see whether a prompt works, this alone changes the economics of iteration.

Fourth, video references. Omni 1.1 accepts up to three seconds of uploaded footage as a style and motion reference, letting creators carry characters, choreography, and motion patterns across generations. Google’s example: replace three reference dancers with a dog, an octopus, and a bear, each performing the specific dance from its assigned reference clip, in one continuous shot.

The pricing math

Per-second pricing for Omni 1.1 Flash: $0.03 at 360p, $0.10 at 720p, $0.15 at 1080p, and $0.30 at 4K. That positions it deliberately across Google’s own lineup — cheaper than Veo 3.1 Fast at 1080p ($0.12 vs. $0.15 is a loss, but 720p parity at $0.10 holds), while 360p drafts undercut everything else on the market. A full 40-second 720p scene runs about $4.00; a 360p draft of the same length costs $1.20. For agencies and studios testing dozens of variations before committing to a hero shot, the draft tier is the real story.

Who’s already shipping with it

Google also used the launch to flex adoption. Adobe has integrated Gemini Omni Flash into Firefly; Figma Weave uses it as a core video engine for its creative canvas, with creative director Itay Schiff noting that extensions, richer references, and 4K “take teams beyond generating videos to truly directing them.” Runway — nominally a Google competitor in generative video — offers Omni Flash alongside its own models, with CCO Jamie Umpherson describing it as fitting “naturally into how people already use Runway.” GMI Cloud cited its accuracy for educational content where details “hold up under scrutiny.”

That last point matters more than the marketing. The persistent complaint about AI video has never been resolution — it’s reliability: hands, text, physics, and continuity breaking under close inspection. A 10-second context window won’t fix every artifact, but it attacks the continuity problem directly, and partners choosing Omni for accuracy-sensitive verticals like education suggests the model holds up better than its predecessors under exactly that kind of scrutiny.

Available everywhere at once

Unlike some staged rollouts, Omni 1.1 landed broadly on day one: in Google AI Studio and via the Gemini API, on the Gemini Enterprise Agent Platform for enterprises, and for all Google AI Plus, Pro, and Ultra subscribers globally in Google Flow, with scene extension also live in the Gemini app. The API itself is minimal — the code sample in Google’s post is a short genai.Client() call specifying the model, a previous interaction ID, and a target resolution. Low integration friction is clearly part of the strategy.

Why this matters

The generative video market is consolidating around a thesis: the winning model isn’t the one with the prettiest single generation, it’s the one that behaves like a controllable production pipeline — drafts you can afford to iterate, keyframes you can direct against, references that preserve identity, and an upscale path to delivery resolution. OpenAI’s Sora, Runway’s Gen-4, and Kling have all been racing toward the same checklist. Omni 1.1 Flash is the most complete version of it shipped so far, and at the most aggressive price points at the draft tier.

The gap between “impressive demo” and “tool a studio can bill against” has always been control, cost, and reliability. Google just moved all three dials at once. The next twelve months of AI video won’t be about whether models can generate beautiful seconds — they already can. It’ll be about who can sustain a coherent 40-second scene at a price a production budget can absorb. As of this week, that race has a new leader.