Gemini Omni 1.1 Flash: generate, reference, edit, and extend video

Use the Omni 1.1 video route with native sound, first/last frames, multiple references, and output up to 4K.

Where: Create Video or a supported video edit/extend surface → Gemini Omni 1.1 Flash.

Key ideas

Video model, not the chat model: The exact ID is gemini-omni-1.1-flash-preview. Gemini 3.8 Flash is a language-model choice and cannot substitute for this route. Legacy Gemini Omni Flash remains a separate model with different limits.
Intent and inputs choose the task: Text-only input generates a clip; images can guide it; multiple reference videos/images use reference generation. One source video on the general generation route lets the provider interpret the prompt, whereas an explicit annotation edit or Extend Video requests that specific task.
Native sound and resolution: This route supports 360p, 720p, 1080p, and 4k output with native sound. The 1.1 profile accepts last frames and continuation; those capabilities must not be assumed for legacy Omni.

Select Gemini Omni 1.1 Flash and describe the new clip or precise requested change.
Connect first/last-frame or reference media using the inputs exposed for the selected mode. Confirm each video is at most 10 seconds.
For a video annotation edit, save the annotation changes before running. For continuation, use Extend Video so the request explicitly asks to extend the source.
Choose a supported output resolution and duration from the current controls. Check the cost after adding references, because source video is charged separately.
Run, review both motion and sound, and keep the successful take or extract it to a reusable input node.

Up to 10 images and 3 input videos, each video up to 10 seconds.
Explicit video edit requires a source clip and cannot simultaneously extend it.
Multiple videos or reference images choose reference-to-video; one general-purpose source video uses the prompt-led auto task.
The older Omni profile accepts up to 30 images, one video, and 720p only; its different limits do not apply to 1.1.

Output rates are 2 / 6 / 9 / 18 Sparks per second at 360p / 720p / 1080p / 4k.
Input video adds 1 Spark per source second.
A reference-image run adds a flat 1 Spark, not 1 Spark for every image.
Example: a 3-second 4K output, 10 seconds of input video, and image references quotes 54 + 10 + 1 = 65 Sparks.

Provider limits are validated before the request. A private or unreadable media URL cannot be used simply by pasting it into a prompt.
General one-video generation is prompt-led; use the explicit edit or extend surface when that distinction matters.

Trim each source below the per-video limit and wait for duration detection. Also check that the selected model is Omni 1.1, not legacy Omni.

Confirm the latest annotations are saved and that the edit source is connected; inspect the chosen source/take before running again.