Google launched Gemini Omni 1.1 Flash on August 27, 2026, as the generally available version of its fast, conversational video-generation and editing model. Although this article is a preview in the editorial “first look” sense, the product itself is no longer a preview: the stable Gemini API model ID is gemini-omni-1.1-flash. (Google release notes)
That distinction matters because the older gemini-omni-flash-preview endpoint is scheduled for deprecation on September 30, 2026. If you have seen the informal search phrase Google omni flash, Gemini Omni 1.1 Flash is the current official model name to use in product copy and API integrations.
What Is Gemini Omni 1.1 Flash?
Gemini Omni 1.1 Flash is a native multimodal model for video generation and editing. “Multimodal” means it can work across different media types rather than treating text, images, audio, and video as isolated inputs. Google positions the model around three ideas: understanding mixed media together, applying real-world knowledge, and editing video through natural-language conversation.
The API produces 3- to 10-second videos with audio at 24 frames per second. It supports landscape 16:9 and portrait 9:16 output, with 720p as the default resolution. The documented context window is 1,048,576 tokens, while current output choices include 360p, 720p, and upscaled 1080p or 4K. For developers, this makes the model less like a one-shot clip generator and more like a video engine that can sit inside an iterative creative product. (Official model page)
What Is New in Gemini Omni 1.1 Flash?
The biggest upgrade is control. Gemini Omni 1.1 Flash can examine up to 10 seconds of prior footage when extending a scene, rather than relying only on the final second as earlier versions did. Extensions are generated in 10-second increments and can reach a cumulative length of 40 seconds. That extra context should help preserve motion, characters, audio, and narrative direction across connected shots. (Google launch announcement)
First-and-last-frame interpolation adds another production-oriented tool. A creator can supply two key images and describe the transition between them, leaving the model to generate a continuous camera move or scene transformation. This is useful for loops, orbits, zoom transitions, and shots with a predetermined ending.
The new resolution workflow also encourages deliberate iteration. Google says 360p drafts can generate up to 60% faster and cost one-third as much as standard 720p output. Once a direction works, users can render at 720p or request upscaled 1080p or 4K delivery. The word “upscaled” is important: Google does not describe those higher-resolution modes as native 1080p or native 4K generation. (Google launch announcement)
Gemini Omni 1.1 Flash Review: Practical Strengths
On paper, the model’s most valuable feature is not resolution; it is conversational continuity. Using the Interactions API, an application can preserve an earlier interaction and request a targeted change such as altering an object or continuing a scene. That creates a familiar draft-review-revise loop instead of forcing creators to rewrite a full prompt for every attempt.
The combination of low-cost drafts, keyframe control, extensions, reference inputs, and final upscaling fits advertising, previsualization, social video, education, and creative-tool workflows. A sensible pipeline would generate several 360p candidates, compare motion and composition, refine one conversationally, and upscale only the approved result. This could reduce wasted high-resolution renders, although teams should still test identity consistency, text rendering, and complex motion on their own content.
Gemini Omni 1.1 Flash Benchmarks
Google publishes preference-based evaluations rather than a single universal quality score. These results use human side-by-side judgments, and Google does not expose numerical win rates on the public model page, so they should be read as vendor-reported directional evidence—not independent proof.
Evaluation | Test scope | Google-reported result |
|---|---|---|
Video editing | 504 examples | Leading overall preference and instruction following |
MovieGenBench text-to-video | 1,003 prompts | Best overall preference and instruction following |
VBench image-to-video | 355 image-text pairs | Tied with Grok Imagine Video and Kling, ahead of other tested models |
Reference-to-video | 468 mixed-reference examples | Leading overall preference and speech adherence |
The takeaway is that Gemini Omni Flash appears strongest where direction-following and cross-media consistency matter. However, the published comparisons do not isolate every 1.1 feature or provide reproducible win-rate percentages. Buyers should treat them as a reason to run a pilot, not as a substitute for testing real prompts and brand assets.
Gemini Omni 1.1 Flash Pricing
As of August 31, 2026, Gemini Omni 1.1 Flash pricing is available only on the paid Gemini API tier. Input across text, image, video, and audio costs $1.50 per million tokens. Text output costs $9.00 per million tokens, while video output costs $17.50 per million tokens. (Official Gemini API pricing)
Video billing depends on resolution-specific token consumption:
Output resolution | Tokens per second | Approx. effective cost per second |
|---|---|---|
360p | 1,931 | $0.034 |
720p | 5,792 | $0.10 |
1080p | 8,688 | $0.15 |
4K | 17,376 | $0.30 |
These estimates explain the value of draft-first production. A 10-second 360p concept costs roughly $0.34 in video-output charges, compared with about $1.01 at 720p or $3.04 at 4K, before input and any text-output charges. There is no free API tier for the model.
Availability and Important Limits
Developers can access Gemini Omni 1.1 Flash through Google AI Studio and the Gemini API, while enterprise teams can use the Gemini Enterprise Agent Platform. Google also makes Omni 1.1 available in Flow for Google AI Plus, Pro, and Ultra subscribers; scene extension is available to those subscribers in the Gemini app.
The model still has meaningful constraints. Google’s model card says complete consistency across edits, complex motion, and perfectly accurate text remain challenging. The current Gemini API does not support uploaded audio references or voice editing. Uploaded-video editing and extension are unavailable in the EEA, Switzerland, and the UK, and extensions can only be appended to the end of a clip. Multi-video reasoning is not supported, while audio in video references is ignored. (Google DeepMind model card; Gemini API limitations)
Gemini Omni 1.1 Flash Verdict
Gemini Omni 1.1 Flash is a credible step from AI clip generation toward controllable video production. Its best ideas are operational: cheap drafts, conversational revision, longer scene continuity, fixed endpoints, and selective upscaling. Creative-software teams and high-volume content producers should test it, especially if their workflow benefits from reference-driven iteration.
The balanced verdict is straightforward: Google omni flash is a useful discovery phrase, but Gemini Omni 1.1 Flash is the official, production-ready model to evaluate. Its Google-reported benchmarks are promising, yet the documented limitations make project-specific testing essential before committing it to client-facing or high-consistency work.
