Guide2026/08/28

How to Extend Gemini Omni 1.1 Flash Videos to 40 Seconds

Extend Gemini Omni 1.1 Flash videos in 10-second steps to 40 seconds. Prompts, API code, cost math, and what to do past the 40-second cap.

How to Extend Gemini Omni 1.1 Flash Videos to 40 Seconds

Quick Answer

Gemini Omni 1.1 Flash scene extension continues a finished clip in 10-second steps, reading the last 10 seconds for continuity, up to 40 seconds total. A single generation is still 3–10 seconds. What changed on August 27, 2026 is that you can now append: tell the model "Continue the scene" (or describe the next beat), and it writes onto the tail instead of starting a new shot. Google's API docs are explicit — some of the final frames of your input are rewritten so the join is seamless, and audio can carry with the picture. This guide is the how-to: working prompts, the two API paths (multi-turn vs uploaded file), cost at every resolution, the hard limits, and what to do when 40 seconds is not enough.

Need a minute-long cut, a YouTube Short with hard cuts, or a product video past 40 seconds? That's a different job — storyboard stitching and last-frame chaining.

Key Takeaways

  • Base clip: 3–10 seconds. Extended clip: up to 40 seconds. Extension is additive, not a longer first render.
  • Context is 10 seconds, not 1. That's the leap vs Veo-style continuation. Characters, motion, and audio hold because the model sees a full beat, not a freeze-frame.
  • Prompt the next action, not "make it longer." "Continue the scene." works. "Extend this video" works. A camera move, a line of dialogue, or "The music continues into the chorus" works better.
  • Draft at 360p ($0.03/s), keep at 720p ($0.10/s), finish via 1080p/4K upscale. A 40-second 720p take is $4.00 raw; the same length at 360p is $1.20.
  • You can only append. No prepend, no mid-clip insert. Uploaded source clips must be ≤10 seconds. Extending uploaded video is currently blocked in the EEA, Switzerland, and the UK — extending a clip the model just generated is not.
  • Past 40 seconds, chain shots. Native extension stops there. The longer-than-10-seconds playbook covers stitching and last-frame chains.

How scene extension works

Google's announcement frames it as "extend a scene": take a finished clip and keep generating from where it left off. The Gemini API docs fill in the mechanics:

  1. You already have a clip — either one you just generated, or a file you uploaded (≤10 seconds).
  2. You send a continuation prompt on the same interaction (or with the uploaded file).
  3. The model reads up to 10 seconds of prior footage, writes a 3–10 second continuation, and rewrites a few tail frames of the source so the cut doesn't flash.
  4. Repeat until the cumulative length hits 40 seconds.

That 10-second lookback is the actual product. Earlier continuation tricks (and Veo's 1-second tail) guessed the next motion from a still. Omni 1.1 Flash sees the walk-up, the glance, the line of dialogue — so the next 10 seconds can stay in the same world.

Two things it is not:

  • Not a 40-second first take. Asking for "a 40-second tracking shot" on a fresh prompt still returns ≤10 seconds. You generate, then extend.
  • Not conversational editing. "Make the jacket red" re-renders the same clip. "Continue the scene: she turns and walks to the door" appends. Same previous_interaction_id mechanism, different intent. See conversational editing for the refine loop.

How to extend a clip (no code)

This is the path in the Gemini app, Google Flow, and AI Studio once 1.1 has reached your account (AI Plus / Pro / Ultra for the consumer apps).

  1. Generate a 3–10 second base clip you actually like. Don't extend a take you'd re-roll — every extra 10 seconds is billed, and a weak start poisons the next three.
  2. End the base clip on a pause. A look, a landing, a held pose. Mid-motion blur at the tail is what the model has to rewrite, and that's where seams show.
  3. Prompt the next beat only. One camera move, one action, one audio note. Examples that match Google's own prompt guide:
    • "Continue the scene."
    • "The man in the blue sweater replies: 'Did your father go out on the boat too?'"
    • "The camera pulls back in one continuous movement as he continues his story. The music swells."
  4. Say whether the shot continues or cuts. "Show the same characters in the next scene" is a cut. "One continuous shot, no jump cuts" is not. The model will do either — you have to pick.
  5. Name the audio if it should change. "The music continues into the chorus." Silence that, and the score often resets.
  6. Repeat up to 40 seconds. Three extensions on a 10-second base is the full tank.

Draft the whole chain at 360p first. 1.1's draft mode is up to 60% faster at one-third the 720p cost — iterate the story cheap, then re-render or upscale the keeper. That's the 360p → 4K loop Google is pushing.

If you'd rather not babysit model strings and interaction IDs, Omni Flash runs an Omni-era generate-and-refine loop in the browser. Native 40-second extension still lives on Google's 1.1 string; the studio is the place to storyboard and chain when you go past that cap.

How to extend via the API

Model ID: gemini-omni-1.1-flash. Full setup (key, SDK, pricing) is in the API tutorial. Two shapes, both from Google's docs.

Path A — extend a clip you just generated (works in every region that has the model). Pass previous_interaction_id and a continuation prompt. Draft at 360p:

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",
    },
)

Path B — extend an uploaded file (source must be ≤10 seconds). Not currently available in the EEA, Switzerland, or the UK:

from google import genai

client = genai.Client()
video_file = client.files.upload(file="clip.mp4")

interaction = client.interactions.create(
    model="gemini-omni-1.1-flash",
    input=[
        {"type": "document", "uri": video_file.uri},
        {"type": "text", "text": "Continue the scene: the camera pans across the mountains."}
    ],
)

JavaScript is the same shape on @google/genai:

const interaction = await ai.interactions.create({
  model: 'gemini-omni-1.1-flash',
  previousInteractionId: previous.id,
  input: [{ type: 'text', text: 'Continue the scene.' }],
  responseFormat: { resolution: '360p' },
});

Loop Path A until you hit 40 seconds. Each call bills for the new seconds, not the whole running clip.

Prompts that actually continue

Google's own video docs list the verbs that trigger extension: "Extend this video" and "The scene continues". Everything else is craft.

GoalPrompt pattern
Hold the shot"Continue the scene." / "Extend this video."
Camera move"Continue the video. Slow pull-out, same lighting, same grade."
Dialogue"She answers: 'I see it too.' Dramatic music score."
Same people, new shot"Show the same characters in the next scene."
Audio carry"The music continues into the chorus."
Timed cut inside the extension"After 2s cut to a new scene with the same characters."

One timestamp trap: 0s means the start of the extension, not the start of the original clip. If you already have 10 seconds and you write "After 2s cut…", the cut lands at 12 seconds on the finished file.

Carry a character reference or a 3-second video reference if a new person has to walk in. Extension keeps what's already on screen; it does not invent a consistent new face from a name in the prompt.

What a 40-second extension costs

Per-second rates from Google's 1.1 ladder: $0.03 at 360p, $0.10 at 720p, $0.15 at 1080p, $0.30 at 4K. You pay for every generated second, including the base clip.

Finished length360p draft720p1080p4K
10 seconds (base only)$0.30$1.00$1.50$3.00
20 seconds (one extension)$0.60$2.00$3.00$6.00
40 seconds (full tank)$1.20$4.00$6.00$12.00

The cheap path is the one Google designed: prototype the whole 40 seconds at 360p ($1.20), then upscale the keeper. Re-rolling a 40-second 4K take is a $12 mistake; re-rolling a 360p draft is $1.20. Conversational-edit turns on the base clip before you extend still bill like generations — polish the 10-second start first.

Where scene extension breaks

  • 40 seconds is a wall, not a soft cap. A fourth extension does not give you 50 seconds. Past 40s, chain shots.
  • End of clip only. You cannot prepend an intro or splice a beat into the middle.
  • Uploaded sources must be ≤10 seconds. Longer files have to be trimmed first. Model-generated multi-turn clips are the exception — that's how you climb to 40.
  • Regional gap on uploads. Extending an uploaded file is blocked in the EEA, Switzerland, and the UK. Extending a clip the model generated in-session is supported in every region the model is in.
  • A bad tail poisons the next clip. Because Omni rewrites the last frames to hide the join, a chaotic or motion-blurred ending becomes the new "truth" the continuation has to match.
  • Audio still isn't a continuous session past the join if you leave the native path. Inside one extended clip, Omni can keep speech and score coherent (Google flags multi-turn voice extension as supported on generated video). The moment you leave 40 seconds and stitch in an editor, treat audio as a separate bed.

FAQ

How long can Gemini Omni 1.1 Flash videos be?

3–10 seconds per generation, extendable in 10-second steps to 40 seconds total. That 40-second figure is cumulative scene length, not a single-request duration.

How do I extend a Gemini Omni 1.1 Flash video?

Generate a clip you like, then send a continuation prompt — "Continue the scene" or a specific next beat — on the same interaction. In the API that's previous_interaction_id plus the new text. In the Gemini app and Flow it's the extend / continue control rolling out with 1.1.

Does "Continue the scene" make the video longer, or does it re-edit it?

It appends. "Make the light warmer" re-edits the current clip at the same length. "Continue the scene" writes new seconds onto the tail. Use the refine wording until the start is right; switch to continue wording when you want length.

Can I extend an uploaded MP4?

Yes, if it's 10 seconds or shorter, and if you're outside the EEA / Switzerland / UK restriction on uploaded-file extension. Trim first. For in-session generations, use the multi-turn path instead — that's how you reach 40 seconds.

How much does a 40-second Omni 1.1 Flash video cost?

$4.00 at 720p, $1.20 at 360p, $6.00 at 1080p, $12.00 at 4K — raw output, before retakes. Draft the chain at 360p, then upscale.

What if I need longer than 40 seconds?

Native scene extension stops at 40. From there, plan shots and chain last frames or stitch a storyboard. That's also the right path for anything that should cut — ads, Shorts, multi-location scenes.

Is Omni Flash affiliated with Google?

No. Omni Flash is an independent platform built for the Gemini Omni era. It is not affiliated with Google. Gemini, Gemini Omni, Omni Flash, Veo, and related names are trademarks of their respective owners.

Next steps