Gemini Omni 1.1 Flash is now available as a production model in the Gemini API for building generative video workflows with explicit controls over resolution, interpolation, editing, and scene extension.
That’s the part that matters.
Not “AI video improved”. That headline is too broad to be useful.
The relevant change is that developers can now target a named model, gemini-omni-1.1-flash, and build on concrete API behaviors: generating drafts at 360p, producing final outputs at 1080p or 4K via upscaling, interpolating between a first and last frame, continuing a scene from an existing clip, and editing generated videos across turns.
Google announced the launch on August 27, 2026 with this phrase: “Today, we’re introducing Gemini Omni 1.1 Flash”. The Gemini API release notes describe it as a generally available model, not just a preview version, and confirm that the earlier gemini-omni-flash-preview endpoint is scheduled to be deprecated on September 30, 2026.
For developers across Iberoamerica building creative tools, internal media workflows, product piece generators, storyboard systems, editing assistants, or audiovisual prototypes, the signal is clear: generative video is starting to look less like a demo box and more like a programmable pipeline.
What is Gemini Omni 1.1 Flash?
Gemini Omni 1.1 Flash is Google’s fast model for conversational video generation and editing from the Gemini API, available with the ID gemini-omni-1.1-flash.
The model accepts text, images, and video as input, and returns video as output. Google’s model page lists support for output videos of 3 to 10 seconds, 24 FPS, and resolutions of 360p, 720p, 1080p, and 4K. That same page describes it with support for video extension, resolution upscaling, and advanced interpolation.
That sets it apart from a simple text-to-video endpoint.
The shape of the API resembles a stateful creative system. You can generate a clip, retain the interaction ID, and request later edits without re-uploading or redescribing everything. You can provide reference images. You can upload a video and request an edit. You can pass two images and ask the model to generate the transition between them.
For those building applications, those details matter more than the quality of an isolated demo. A workflow needs handles. It needs parameters. It needs repeatable primitives that can be turned into product interface.
Omni 1.1 Flash delivers more of those handles.
How to Use Gemini Omni Flash API?
You can use Gemini Omni 1.1 Flash from the Gemini Interactions API by calling the model gemini-omni-1.1-flash.
The official documentation shows examples in Python, JavaScript, and REST. The simplest version is a text prompt with video response format. To control resolution, the documentation shows response_format with a resolution like 1080p.
The API controls worth looking at first are:
model: "gemini-omni-1.1-flash"response_formatresolutionprevious_interaction_idfor stateful editing- image inputs for image-to-video and interpolation
- videos uploaded via Files API
- optional task hints like
image_to_video,edit, orextend
Google’s documentation recommends relying first on the prompt and using the task parameter only when the model needs help choosing the correct mode, because setting task adds constraints.
That’s a useful product design signal.
If you’re building a tool for developers or creators, you probably don’t want every user action to become a rigid task selector. A good interface can expose presets like “extend this clip”, “create a transition”, or “generate a draft”, while letting the prompt carry the creative intent.
What Changes with 360p and 4K Video in Gemini Omni Flash?
The workflow change is that Gemini Omni 1.1 Flash separates iteration quality from final quality.
Google says that 360p previews can be generated up to 60% faster than 720p and at a third of the cost compared to the standard 720p resolution of Omni 1.1. The default resolution is 720p, while 1080p and 4K are available as outputs generated via upscaling.
That’s not just a pricing detail. It changes how a video product can feel.
A serious generative video workflow almost never gets it right on the first result. Users test prompts, camera movement, lighting, timing, subject behavior, references, brand constraints, and scene continuity. If each attempt is slow and expensive, the product becomes cautious. If drafts are faster and cheaper, the product can encourage exploration.
A well-designed application around this probably wouldn’t start with “generate a perfect 4K clip”.
It would start with draft cycles:
- Generate three or four variants at 360p.
- Compare them side by side.
- Keep the most promising direction.
- Refine the prompt or reference media.
- Extend or edit the chosen result.
- Upscale resolution only when the creative direction is stable.
That’s the angle for developers. Resolution isn’t just a render option. It becomes part of the product workflow.
How Does AI Video Interpolation Work?
First and last frame interpolation lets you pass two images to Gemini Omni Flash and ask it to generate the video transition between them.
Google’s documentation describes the pattern straightforwardly: you provide two images in the input list, describe the desired transition, and the model animates from the first frame to the final frame.
This is useful because many real creative workflows already think in keyframes.
A designer might have a product shot at the start and a desired composition at the end. A marketing team might want a brand reveal that starts on an object and ends in a frame with a logo. A video editor might need smooth camera movement between two visual states. An ecommerce or games team might want a view that rotates or transforms an object without losing its identity.
Without interpolation, the user expects a prompt to nail both the start and the end. With interpolation, the application can let the user fix both endpoints and ask the model to solve the movement between them.
That makes generative video a more directed form of multimedia generation.
It’s not deterministic, of course. But it’s much more controllable than relying on text alone.
How to Extend Video with AI Using Gemini Omni Flash?
Scene extension lets Gemini Omni 1.1 Flash continue an existing video by generating a new continuation at the end of the clip.
Google’s announcement says that Omni 1.1 can analyze up to 10 seconds of prior context and extend videos in 10-second increments up to a cumulative total duration of 40 seconds. The API documentation describes extension as a 3 to 10-second continuation from the end of an existing clip.
There are two practical paths.
If the video was generated by the model, you can continue the flow with multi-turn editing using previous_interaction_id. If the video comes from outside the current interaction, you can upload it via Files API and pass it along with an extension prompt.
That makes extension useful for more than “make the clip longer”.
A developer can build branching storyboards, iterative ad concepts, product walkthroughs, lesson visuals, real estate previews, game cinematics, or training clips where the user keeps guiding the next segment. The unit of work stops being an isolated prompt and becomes a sequence of controlled generations.The warning matters: the extension only adds content at the end. Google’s documentation says you can’t prepend content or extend the middle of a clip. It also says videos uploaded for extension must be 10 seconds or less, except in multi-turn flows. And at the time of writing this note, editing or extending uploaded videos is not available to users in the European Economic Area, Switzerland, and the United Kingdom, although editing or extending videos generated by the model is supported.
That should shape how the feature is presented in an interface. “Continue this scene” is correct. “Edit any part of any video” is not.
What to review before shipping Gemini Omni Flash to production?
Before shipping Gemini Omni 1.1 Flash to production, it’s worth treating it as a ready-to-use API model, but with time-sensitive operational details.
As of August 28, 2026, the model ID in the documentation is gemini-omni-1.1-flash. The release notes say the earlier preliminary endpoint, gemini-omni-flash-preview, will be deprecated on September 30, 2026. The documentation also surfaces several behaviors that matter for product architecture.
First, large video outputs need careful retrieval. The documentation recommends URI delivery for videos larger than 4 MB, especially above 720p when available. That means a production app should think about async handling, polling, failure states, storage, retries, and visible progress for the user.
Second, stateful editing depends on storage decisions. The documentation notes that using store=false can improve faster synchronous generation, but it also prevents editing the generated video later via previous_interaction_id. That’s a product decision, not just an optimization.
Third, regional restrictions matter. Editing and extending uploaded videos are not available in all markets. If your user base includes Europe, Switzerland, or the United Kingdom, the interface and documentation need to distinguish between workflows with generated videos and workflows with uploaded videos.
Fourth, pricing needs review before shipping a calculator or committing margins. Google’s announcement gives the draft comparison at 360p, but pricing surfaces and versions can move fast around newly-GA multimedia models.
Finally, this is still generative media. Developers need review flows, moderation limits, rights validation for uploaded media, and clear labels to explain what was generated, edited, extended, or upscaled.
The model gives you more control. It doesn’t remove the responsibility to product.
Why this matters for AI developer tools?
Gemini Omni 1.1 Flash matters because video generation is starting to surface the controls developers need to build real tools.
The first wave of AI video products was mostly prompt boxes and galleries. That works for demos, but falls short for software. Developers need state, references, drafts, final renders, explicit output configuration, specific modes, file handling, and constraints that can map to interface decisions.
Omni 1.1 Flash moves in that direction.
It gives a product team enough structure to build a real creative workflow:
- A draft room using 360p.
- A final render path using 1080p or 4K.
- A keyframe transition tool using interpolation.
- A function to continue scenes using extension.
- A conversational editor using
previous_interaction_id. - A file pipeline for uploaded reference videos.
- A generation history that can preserve or discard editability based on what the product needs.
That’s why this launch matters for developers.
The market headline will be “Google’s AI video got better.” The technical headline is more precise: Google’s AI video model is becoming an API surface you can build products on.
That’s the signal worth following.
It also fits a broader Google line in developer tools. At yoDEV we’ve already seen how Gemini 2.0 Flash targets cost and performance of models for agentic work, how Gemini 2.0 Ultra pushes context and native execution, and how Antigravity 2.0 shows the API can matter more than the surface demo.
With Omni 1.1 Flash, that same logic comes to video.