Gemini Omni Flash 1.1: a new model for AI video

Gemini Omni Flash 1.1: What Google’s New AI Video Model Can Do

On August 27, 2026, Google moved its video generation and editing model, Gemini Omni Flash 1.1, out of experimental status. This update to the previous Gemini Omni Flash 1.0 introduces significantly more control over the output, from extending scenes to specifying the first and last frames. The model is already available in Sibrik AI under the name gemini omni flash. Let’s explore what’s new and how to use these features in practice.

How Gemini Omni Flash 1.1 Differs from Version 1.0

The main difference is that the model has moved from experimental status to production, bringing several practical tools for controlling the output:

• Scene extension. The model can now analyze up to 10 seconds of the preceding video instead of just the last second, as in the previous version. This makes it possible to extend a clip in 10-second increments, producing up to 40 seconds of video from a single source frame or clip.
• First and last frame control. You can specify the starting and ending frames, and the model will generate a smooth transition between them. This is useful for complex camera movements, zooms, and seamless loops.
• Flexible duration and upscaling. Available resolutions include 360p, 720p, 1080p, and 4K. You can generate lightweight 360p drafts and upscale the final version to a higher resolution.
• Reference videos for consistency. Short video clips, up to a few seconds long, can be used as visual references to maintain a character’s appearance or scene style across frames.
• Built-in audio. Audio is generated in sync with the video in the same prompt, without requiring a separate generation pass.

Gemini Omni Flash 1.1 in Sibrik AI: Working Modes

In Sibrik AI, the model is available under the identifier gemini omni flash and supports two workflows:

• Text-to-video — the model generates a clip from scratch based on a text prompt, without requiring a source video.
• Video-to-video — an existing video is provided as input, and the model transforms, extends, or edits it according to the prompt.

Supported durations are 4, 6, 8, and 10 seconds, with resolutions ranging from 360p to 4K. Audio is generated alongside the video. You can also specify the first and last frames, allowing the model to calculate a smooth transition between them.

When to Use Scene Extension vs. Frame Control

Here’s a quick guide to choosing the right approach for your task:

Scene extension — when you already have a short clip, for example, from Kling or Seedance, and want to continue the action without starting the generation process from scratch.

First and last frame control — when a specific result is important at the beginning and end of the video. For example, a product needs to land in a precisely defined position in the final frame of an advertisement.

Reference video — when you need to preserve a character’s appearance or scene style in a new clip without describing every detail in words.

360p draft → upscale — when testing several prompt variations and wanting to preview the composition quickly before generating the final version at a higher resolution.

Here’s an example of a prompt that makes good use of frame control:

“The camera slowly moves toward a coffee cup on a table beside a window. Soft morning light fills the scene as steam rises from the drink. The camera completes its movement with a smooth transition into a close-up of the logo on the cup.”

Combined with specified first and last frames, this approach produces a more predictable, controlled result instead of a randomly composed scene.

Best Use Cases

In practice, Gemini Omni Flash 1.1’s controllability is especially valuable in three scenarios.

First, continuing a story. When a generated 5–10-second clip ends just as the action should continue, scene extension lets you build the video out to 40 seconds without restarting the entire process or losing the composition of the preceding footage.

Second, advertisements requiring precise framing. When a product, logo, or text banner must appear in a specific position in the final frame, first and last frame control reduces the unpredictability often associated with standard text-to-video generation.

Third, video series featuring the same character or object. Reference videos help maintain a consistent appearance across multiple separate clips, which is particularly useful for serialized social media content or educational videos featuring the same presenter.

Known Limitations

Based on early user feedback, there are several caveats to keep in mind:

• Upscaling to 1080p and 4K enhances the base render rather than generating additional detail from scratch, so the improvement in sharpness is moderate.
• The 360p preview is rendered separately from higher-quality versions, meaning it does not always accurately represent the final frame.
• The model sometimes interprets action descriptions loosely. Complex or highly specific movements should be checked, and prompts refined when necessary.
• Small text in the frame may lose clarity when scene lighting changes.
• First/last frame control and a set of reference images cannot be used in the same request, so you must choose one approach per generation.

FAQ

How does Gemini Omni Flash 1.1 differ from standard text-to-video models like Kling or Seedance?

The main difference is its focus on controllability and editing: extending existing scenes, specifying the first and last frames, and working with reference videos. It is more of a tool for precisely refining and extending video than for generating complex narratives from scratch.

Can I extend a video generated with a model other than Gemini Omni?

Yes. The model accepts source videos through its video-to-video mode and can extend or transform them according to a new prompt, regardless of which model generated the original footage.

Does the model support vertical video for Reels and TikTok?

Yes. It supports 16:9 and 9:16 aspect ratios, making it suitable for both horizontal and vertical formats.

Conclusion

Gemini Omni Flash 1.1 marks a shift from an experimental model to a fully fledged tool for AI video creation. Its features include scene extension up to 40 seconds, precise first and last frame control, reference videos for consistency, and resolution options ranging from 360p drafts to 4K output.

For tasks where controlling the final result matters as much as generating content, it is currently one of the most flexible options in Google’s model lineup.

Try Gemini Omni Flash 1.1 now in Sibrik AI. Explore the full catalog of video, image, and music models at sibrik.ai/models.