Gemini Omni 1.1 Flash: 4K Output, 360p Drafts at 1/3 Cost

Google DeepMind shipped Gemini Omni 1.1 Flash on August 27, 2026, positioning it as a production-ready successor to the original Omni Flash with developer control as the central design goal. Four discrete capabilities arrive together — scene extension, first/last-frame interpolation, resolution-tiered draft generation, and 4K upscaling — each exposed directly through the Gemini API. For engineers building generative video pipelines, that control ladder matters as much as raw quality improvements.

Scene Extension

The most architecturally significant change is context depth during scene extension. The prior model referenced only the final second of an existing clip; Omni 1.1 Flash analyzes up to 10 seconds of prior footage. That tenfold increase in lookback reduces visual drift and narrative inconsistency across extended sequences. The model generates extensions in 10-second increments and supports cumulative clip lengths of up to 40 seconds.

The API surface is straightforward: pass a previous_interaction_id to chain generations, optionally setting resolution in response_format at each step. This chaining model puts extension logic in the application layer rather than requiring a single long inference call — a pattern consistent with the pipeline architecture thinking that increasingly drives AI gains in 2026.

Keyframe Interpolation and Video Reference Input

Omni 1.1 Flash accepts both a starting and an ending frame, generating the interpolated video between them. Declared use cases — camera orbits, zoom transitions, seamless loops — all require the model to maintain subject identity and scene geometry across the generated span, where the extended context window pays additional dividends. Separately, the model accepts up to three seconds of video as reference input alongside static images and text prompts, enabling character and visual-style consistency across generations without fine-tuning.

Resolution Tiers as a Cost and Latency Control

The clearest developer-facing pricing lever is the 360p draft tier. Google reports up to 60% faster generation throughput at 360p compared to standard 720p output, with cost at one-third of the 720p rate. The intended workflow is explicit: iterate at 360p, promote final takes to 1080p or 4K. This maps latency and spend against fidelity in a way that fits storyboard and approval workflows directly. Figma Weave's Creative Director specifically called out the ability to "branch different versions" as the key production behaviour enabled by these controls.

Availability

Omni 1.1 Flash is live in Google AI Studio and accessible via the Gemini API under the model identifier gemini-omni-1.1-flash. Enterprise deployments can route through the Gemini Enterprise Agent Platform. Consumer access is available in Google Flow for Google AI Plus, Pro, and Ultra subscribers globally; scene extension is also available in the Gemini app for those same tiers. Adobe Firefly, Figma Weave, GMI Cloud, and Runway have all confirmed production integrations. Google has published a dedicated pricing table in the announcement; per-unit figures are not reproduced in the source text and should be confirmed in the official documentation.

The 1.1 Flash release demonstrates that meaningful differentiation in generative video is shifting from capability existence to developer control over what is generated and at what cost. The tiered resolution system is the kind of systems-engineering lever that now rivals raw scaling as a source of capability gains — the same model, exposed differently, becomes usable in workflows that were previously too expensive to prototype. For teams evaluating Gemini API investment, the 360p-to-4K pipeline pattern and the chained previous_interaction_id mechanism are the two implementation details worth prototyping against first.