Gemini Omni 1.1 Flash lets you build with more control

What it claims

Gemini Omni 1.1 Flash (API model id gemini-omni-1.1-flash) is Google DeepMind's update to its generative video model, pitched as "production-ready for professional use" via the Gemini API. The default output resolution is 720p. The page adds three specifics the summary leaves out:

• Scene extension context. Extension now conditions on up to 10 seconds of prior video. Previous models referenced only the final second.
• Draft mode cost. 360p drafts cost one third as much as 720p and generate up to 60% faster.
• Video reference length. Video references are capped at three seconds of input per reference.

How to use it

Extension chains off a prior generation by its interaction ID. It does not take an uploaded file. The call goes through client.interactions.create in the google-genai SDK:

from google import genai
client = genai.Client()
interaction = client.interactions.create(
    model="gemini-omni-1.1-flash",
    previous_interaction_id=previous_video_interaction.id,
    input=[
        {"type": "text", "text": "Continue the scene."}
    ],
    response_format={
        "resolution": "360p",
    },
)

Resolution is set through response_format. The page does not show request shapes for keyframes, upscaling or video references. It points to the official docs, cookbook and prompting guides for those.

The multimodal reference example mixes several inputs in one prompt: still images (dog.png, octo.png, bear.png), dance videos (dance1-3.mp4) and a background image. The prompt assigns each character to a specific video by filename.

Prompts for the keyframe and reference examples repeat "One continuous shot, no jump cuts." This suggests the phrase is part of how Google steers the model toward continuity.

Availability:

• Google AI Studio (Gemini API).
• The Gemini Enterprise Agent Platform API, which is the route the named production customers use.
• Google Flow, for Google AI Plus, Pro and Ultra subscribers.
• The Gemini app offers scene extension only, for the same subscriber tiers.

Limits

• Length cap. Extension works in fixed 10-second increments, with a hard cap of 40 seconds total.
• Speed figure. The "up to 60% faster" claim is based on system throughput of 360p vs 720p. It is not a per-request latency figure.
• No published pricing. The page refers to a pricing table, but no prices appear in the text. Absolute per-second cost has to be checked separately.
• No quality evaluation. There are no benchmarks or quantitative measures for consistency, prompt adherence or upscaling. Evidence is limited to example prompts and showcase videos.
• Upscaling unclear. The page does not say whether 1080p/4K output is a separate upscale call or a direct generation option, or how either is priced.