Gemini Omni 1.1 Flash Lets You Build with More Control — Scene Extension, Frame Control, and 4K Arrive

Gemini Omni 1.1 Flash shipped on August 27, 2026 as a production update to the multimodal video model, turning a 10-second single-shot generator into something closer to a controllable editing instrument. The headline numbers are easy to miss because the price did not move: a 720p clip still costs $0.10 per second. What changed is how much of the shot you get to decide — the model now extends scenes up to 40 seconds, honors a specified first and last frame, drafts previews at 360p for about a third of the 720p cost, and renders final output at 1080p or 4K. It is the kind of update where the release notes matter more than the launch.

The model is the successor to the Gemini Omni Flash preview that became available to developers on June 30, 2026, and it keeps everything that made the original distinctive: multimodal input — text, images, audio, and video in a single call — and conversational, multi-turn editing where you refine a clip by talking about it rather than regenerating from scratch. What Gemini Omni 1.1 Flash adds is control at the boundaries and control over cost, which is precisely what a production pipeline needs and a research demo does not.

What the update actually adds

Four capabilities separate 1.1 from the June preview, and each answers a concrete complaint builders had with the original.

• Scene extension. The model can now analyze up to 10 seconds of prior video context when continuing a clip — earlier versions of the model referenced only the final second. Extensions happen in 10-second increments, up to a 40-second cumulative total. For the first time you can produce something that reads as a sequence with continuity instead of a set of adjacent clips.

• First/last frame control. Specify the starting and ending frame of a shot and the model generates continuous video between the two keyframes. This is what makes camera orbits, zoom transitions, and seamless loops feasible without a video editor, because the geometry of the shot is now a constraint instead of a guess.

• 360p drafting mode. Lightweight previews render up to 60% faster than standard 720p, at roughly a third of the cost. That is a storyboard/iteration tool: you test a dozen variations cheaply, then render only the one that works at full resolution.

• 1080p and 4K output. Final renders can be upscaled from the original 720p ceiling to 1080p or 4K, which is the difference between "usable for social" and "usable for broadcast." Video references up to three seconds long can now also be included as multimodal input to hold character and visual consistency across shots.

The pricing picture is still settling

The $0.10-per-second figure that everyone quotes is the 720p standard tier, and it is unchanged from the preview. Input is $1.50 per million tokens whether those tokens are text, images, audio, or video, and text output is $9.00 per million tokens — the same structure as the original. What Google has not yet published is per-second pricing for the new 1080p and 4K tiers; third-party estimates put them in the 1.5x to 3x range of the 720p base, but those are estimates, not Google's numbers, so treat any production budget at high resolution as provisional until the official rate appears on the Gemini API pricing page.

The 360p drafting tier is the deliberate cost lever: at roughly a third of the 720p price and up to 60% faster, it changes the economics of iteration. A team doing ten drafts of a shot at 360p spends roughly the same as three drafts at 720p, which is the difference between exploring and committing early. Vendor-reported throughout, but the throughput math is simple arithmetic on Google's own published figures.

Where you can actually call it

Gemini Omni 1.1 Flash is available now through the Gemini API in Google AI Studio under the model ID gemini-omni-1.1-flash, and through the Gemini Enterprise Agent Platform for production workloads — the same interface that Adobe, Figma, GMI Cloud, and Runway are already building on for their Agent Platform integrations. Consumer access rolled out in parallel: Google Flow has the full feature set for all Google AI Plus, Pro, and Ultra subscribers globally, and scene extension is live in the Gemini app. In other words, this is not a paper launch — the API, the agent platform, and the consumer surfaces all moved on the same day.

That breadth is worth a moment. The original Gemini Omni Flash shipped to developers on June 30 with a public-preview label, a hard 10-second generation limit, scene extension explicitly unsupported, and video references accepted by the schema but not correctly processed. 1.1 clears most of that list in one update — the 40-second ceiling, working video references, and a proper resolution ladder are exactly the gaps the preview era left open. If you bounced off the preview because it could not hold a scene together past ten seconds, this release is the answer to that specific complaint.

What has not changed

For all the new control, the underlying product is the same model family with the same trade-offs. Output is still short-form video with synchronized audio — this is not Veo 3.1, Google's separate one-shot high-fidelity generator, and the two are aimed at different jobs. Character consistency across scene changes is improved by the longer lookback and the reference-video input, but it is still a generative weakness you should test rather than assume. And the model carries SynthID watermarking on output, which matters if you are building a tool that needs provenance guarantees.

It is also worth stating plainly what has not been proven yet: there is no independent benchmark for Gemini Omni 1.1 Flash on a public video leaderboard as of this writing. Google's claims — the 60% speedup for 360p, the quality at 1080p/4K — are vendor-reported. The story of this release is the feature list and the price, not a score on a chart, and anyone who tells you different is citing a number nobody outside the lab has produced.

Why "more control" is the story

The through-line of this release is that Google finally gave builders handles: length, framing, resolution, and cost are all inputs now instead of outputs. That is what makes a generative model feel like a tool rather than a slot machine, and it is the same reason cost transparency matters on the infrastructure side. Every generative-model team we talk to is fighting the same two problems — unit economics and iteration velocity — and a model that costs a third as much to iterate on attacks both.

That cost-control theme is where a routing layer earns its keep. OrcaRouter passes provider list prices through with zero markup, so when a model's per-second or per-token rate changes, the price you see on our side changes the same day — no margin between you and Google's published number. One API covers 200+ models, with automatic failover if a provider is down or a day-old model returns errors mid-production; for a model that shipped yesterday, failover is the way to test it without betting a live pipeline on it. Gemini Omni 1.1 Flash itself is not routable through us yet — we do not claim to host a model we do not — but the pricing discipline is the point: when its API tier opens to third-party routing, it will appear at Google's list price, and the integration is a config change, not a rewrite.

Bottom line

Gemini Omni 1.1 Flash is the update that makes Gemini's video model usable as a production tool rather than a novelty: scene extension to 40 seconds, first/last frame control, a 360p drafting tier at a third of the cost, and 1080p/4K output. The $0.10-per-second 720p price is unchanged, high-resolution pricing is still unpublished, and there is no independent benchmark yet — treat Google's feature claims as vendor-reported until someone else runs the model. If you were waiting on the preview because ten seconds was too short and 720p too small, the wait is over. If you need a benchmark or an official 4K rate before you commit, the honest answer is that nobody has them yet.

Watch the Gemini API pricing page for the 1080p/4K per-second rates, watch for the first independent evaluations of the 1.1 update, and watch Google Flow's subscriber rollout as a proxy for how much confidence the team has in the model's quality at scale. The release is real, the controls are real, and the pricing gap is the only part that is still moving.