Gemini Omni Flash launched with a simple proposition: combine text, images, audio, and video, then keep refining the output through conversation. Gemini Omni 1.1 Flash keeps that foundation and tackles the harder production questions. How do you hold a scene together, define where a shot ends, test ideas cheaply, and deliver at a usable resolution?
Released on August 27, 2026, Gemini Omni 1.1 Flash is the stable API model gemini-omni-1.1-flash. Its predecessor, gemini-omni-flash-preview, entered public preview on June 30 after Gemini Omni’s I/O 2026 debut. Version 1.1 carries the preview into general availability with tighter control over continuity, keyframes, resolution, and iteration.
Gemini Omni Flash vs. Gemini Omni 1.1 Flash
Area | Gemini Omni Flash preview | Gemini Omni 1.1 Flash |
API status | Public preview | Generally available, production-ready release |
Model ID |
|
|
Output resolution | 720p | 360p, 720p, upscaled 1080p, or upscaled 4K |
Clip generation | 3–10 seconds | 3–10 seconds, plus extensions up to 40 seconds total |
Keyframe control | No documented first-and-last-frame mode | First-and-last-frame interpolation |
Dedicated video-reference mode | Not documented | Up to three clips, each up to three seconds |
Lifecycle | Scheduled to shut down September 30, 2026 | Stable replacement; no shutdown date announced |
What Hasn’t Changed
At Google I/O, Google described Gemini Omni as a model family that could create from multiple input types, beginning with video. The first Omni Flash combined Gemini’s world knowledge with generative media and supported conversational editing: a user could generate a clip, request a background or style change, and continue refining the same creation. Version 1.1 retains those basics. Google’s model card covers both versions and describes a natively multimodal model trained on text, image, audio, and video data. The API still produces 3–10-second clips at 24 frames per second and has a 1,048,576-token context window.
Longer Scenes with More Context
The clearest upgrade is video extension. Gemini Omni 1.1 Flash can continue a scene in 10-second increments until the complete sequence reaches 40 seconds. According to Google, the model can examine up to 10 seconds of preceding context when generating the continuation; earlier models referenced only the final second.
That extra context is designed to preserve motion, characters, audio, and narrative direction across transitions. In a typical workflow, a creator could approve a 10-second shot, extend it, and review the next segment before continuing. The API only appends footage to the end of a clip, however, and uploaded videos for extension must generally be 10 seconds or shorter.
First-and-Last-Frame Control
The preview model could animate a still image, but Omni 1.1 adds a more direct form of shot planning: creators can supply a starting image and an ending image, then ask the model to generate the movement between them. Google positions this feature for camera orbits, zoom transitions, seamless loops, and other shots where the destination matters as much as the opening composition. With prompt-only generation, the model invents both the route and the destination. Keyframe interpolation fixes two visual anchors and leaves only the movement between them open. For a looping product shot or a transition between two approved compositions, that removes a major source of guesswork.
Faster Drafting and Higher-Resolution Delivery
The original API preview generated 720p video. Omni 1.1 introduces explicit resolution control: 360p for drafts, 720p by default, and 1080p or 4K for final output. Google says 360p generation can be up to 60% faster and cost a third as much as the model’s standard 720p mode. A team can test several camera directions at 360p, choose one, and reserve the more expensive resolution for delivery.
The resolution claim needs one qualification. Google’s developer documentation labels the 1080p and 4K options as upscaled output, rather than native high-resolution generation. That distinction matters when teams evaluate fine detail or plan further post-production.
Video References Add Another Layer of Direction
Omni 1.1 can also use short videos as visual references when creating a new scene. Google’s launch post highlights references of up to three seconds, while the API documentation specifies a maximum of three reference clips, each up to three seconds long. The intended use is narrow but useful: carry a subject’s likeness or a movement pattern into a different composition.
This is not a general “learn from any video” feature. The API ignores audio inside a video reference and does not currently support uploaded audio references or voice editing. Google also warns that reasoning across multiple videos is unsupported. Reference clips should therefore guide likeness or motion, not serve as source material for a multi-video understanding task.
Is Gemini Omni 1.1 Flash Worth the Upgrade?
Developers using the preview endpoint should migrate. Google identifies Gemini Omni 1.1 Flash as its replacement, and gemini-omni-flash-preview is scheduled to shut down on September 30, 2026. The stable model also adds extension, keyframe interpolation, variable resolution, and dedicated video references.
For creators, the upgrade is most useful when a project requires repeated iteration or continuity across shots. Teams producing only isolated 720p clips may notice less of a workflow change. Google still acknowledges problems with complete consistency, complex motion, and perfectly accurate text, so every result needs human review. Omni 1.1 makes the model easier to direct and deploy; it does not eliminate the familiar failure modes of generative video.