Omni 1.1 extends to 40 seconds
Omni 1.1 Flash extends in 10s steps using up to 10s of prior context, to 40s cumulative. That is not Veo scene extension.
Omni 1.1 Flash scene extension is a ten-second continuation that can see up to ten seconds of what already happened, stacked until the file is forty seconds long. It is not Veo scene extension. Google’s 27 August 2026 Keyword post is the source: prior Omni extend used the last one second; 1.1 uses up to ten seconds of prior context; you extend in ten-second increments to a cumulative forty seconds. Versely does not publish a gemini-omni-1-1-flash slug. The Omni rows we do publish stop at ten seconds on generate. If you needed a native thirty-second single pass on this bill, that is Seedance 2.5, not an Omni extend.
This page is only the extend contract. First-last, 360p drafts, and the access map are other posts.
Ten seconds of memory, then ten more seconds of picture
Extension is not “make this clip longer in the abstract.” It is: the model looks at footage you already have, then writes the next beat as if the camera never cut.
The June Omni Flash extend was near-sighted. Google says it referenced the final second. That is enough to match the last color and the last pose. It is not enough to remember that the speaker turned left at 0:04, that the practical on the desk was warm, that the line already said “I see it too.” When the next chunk only sees one second, the third continuation is a new short film that happens to start on a similar frame.
1.1’s change is the lookback. Up to ten seconds of prior context go into the next ten-second write. Google’s claim is improved visual consistency and narrative adherence — the jacket stays the jacket, the room stays the room, the story can continue instead of resetting. You still prompt the continuation. “Continue the scene” is a legal call. “Camera slowly pulls out, forgotten dusty catacombs, dramatic music score” is a directed call. Google’s own examples in the Keyword post are directed: optical dolly-zoom, snap-zoom into the eyes, 360-degree orbit around a frozen beat, a spoken reply. The model is not a hold button. It is a next-shot writer with a ten-second memory.
The increment is ten seconds. The cap is forty seconds cumulative. Four extensions of ten, or a first clip plus three, depending how you count the original — Google’s language is “extend videos in 10-second increments up to a total cumulative length of 40 seconds.” Plan the story as four beats that can survive a handoff, not as one unbroken thought you hope the cap will honor.
This control lives on Google’s 1.1 surface: AI Studio, Agent Platform, Flow for Plus / Pro / Ultra, and scene extension in the Gemini app. It does not live on Gemini Omni Video, which publishes 4s, 6s, 8s, and 10s. It does not live on Gemini Omni Flash Edit, which retouches an existing take from 50 credits and does not promise a forty-second timeline.
Why this is not Veo extend
Google ships more than one way to make a clip longer. Teams collapse them because both buttons say “extend” and both logos say Google.
VEO 3.1 Extend Video is a Veo family row: extend-video category, native audio on, listed duration 7s, Veo 3.1 lineage. You extend a Veo take the way Veo extends: a short continuation of a cinematic clip that already has Veo’s audio contract. You do not get Omni’s ten-second lookback or a forty-second Omni story. You get another Veo beat.
Omni 1.1 extend is Omni’s next ten seconds with ten seconds of Omni memory, capped at forty. Audio, look, and failure modes follow Omni, not Veo. Veo 3.1 generate itself is still 4 / 6 / 8 seconds from 64 credits with a 4K tier. Stacking Veo generates in an editor is a third method and not “scene extension” in either vendor sense.
A fourth method sits next to both: do not extend at all. Seedance 2.5 is a native 4–30 second single pass at 720p, from 34 credits (was 67), native audio. One generate, one continuous action, no handoff at 0:10. If the job is “hold this pour for twenty-two seconds,” 2.5 is the row. If the job is “this Omni scene must keep going, and I am already on Google’s 1.1 door,” Omni extend is the control. If the job is “this Veo take needs seven more seconds of Veo,” Veo extend is the control.
Wrong door, typical waste:
- You extend a Veo clip with Omni 1.1 and the face, the grain, and the room tone change families at the cut.
- You extend Omni Video’s 10s card in your head to 40s, confirm, and receive 10s because the Versely card has no 1.1 extend.
- You buy four Omni 10s generates and call the concatenate “scene extension.” Concatenate is an edit. Extension is conditioned on prior frames.
The AI video generator will not hide this. Duration menus are the contract. If forty is not on the menu, forty is not the SKU.
How to brief a continuation that will actually continue
Write the next ten seconds as a camera operator could execute without a reset.
Lock identity before you extend. If the first clip already drifted the face, ten seconds of memory will faithfully remember the drift. Fix the take on Gemini Omni Flash Edit first, then extend the corrected file on 1.1. Edit is the retouch. Extend is the sequel. Do not ask extend to also recast.
Name the camera move for the next increment, not the whole forty seconds. Google’s examples are one move per call: pan to the man with curly hair; pull out to catacombs; dolly-zoom; orbit. A prompt that lists four locations for one ten-second write will spend the memory budget on a tour.
Write the sound you want in this chunk. Omni 1.1 is a generative video update with audio in Google’s demos (dialogue, score). A silent prompt is not a guarantee of a silent file. If the next ten seconds must be dialogue-free so you can lay Cartesia Sonic 3.5 later, say so. If someone speaks, put the line in the prompt. Native audio on other rows is documented in the native audio glossary; treat Omni extend the same way: assume a track unless you forbade it.
Do not change wardrobe, time of day, or lens family at a handoff unless that change is the story. Ten seconds of context will try to honor the last ten seconds. A sudden night exterior is a new generate wearing an extend label.
Stop at the cap. Forty seconds cumulative is Google’s number, 27 August 2026. A forty-first second is a different clip. Plan the end card inside the last increment.
Caption after the timeline exists, not inside the extend prompt. Add captions to video is the editor job. Extend will not typeset safe-area type.
When to refuse the extend and pick another row
Refuse Omni 1.1 extend when the length is the product, not the continuation.
A product demo that must be one continuous action for twenty-five seconds, no family change, no handoff risk: Seedance 2.5. 4–30s, 720p, from 34 credits, audio on. Brief one verb.
A talking-head that must land a specific line at Veo’s cinematic spec: Veo 3.1, 4 / 6 / 8s, 4K tier, from 64 credits, native audio. Extend that with Veo extend if you must, not with Omni.
A clip that is already good except for the jacket: Omni Flash Edit, from 50 credits. Extending a wrong jacket forty seconds is expensive loyalty.
A first-to-last orbit between two approved stills: that is Omni 1.1 first-last on Google’s door, or VEO First Last Frame on this bill. It is not scene extension. Extension continues time. First-last interpolates two pictures.
A 6-second hook you have not locked: generate on Omni Video at 4–10s, from 29 credits. Do not extend a draft. Google’s 360p draft loop exists so you stop doing that at 720p; use it on 1.1, then extend only the keeper.
If you are on Versely and the job is a short Omni generate or an Omni edit, stay. If you are on Versely and the job is a forty-second Omni memory chain, you are on Google’s 1.1 surface for that chain. Bring the file back. Caption it. Label it if Article 50 applies. The studio is the roster, not the extend button.
Run the Omni rows we publish from the generator, or sign up and pick Edit versus Omni Video by the job. Do not wait for a 40s Omni SKU that is not in the catalog.
FAQ
Is Omni 1.1 extend the same as Veo 3.1 Extend Video?
No. Omni 1.1 extend uses up to ten seconds of prior Omni context and continues in ten-second steps to forty seconds cumulative (Google, 27 August 2026). VEO 3.1 Extend Video is a Veo row: native audio, 7s listed duration, Veo family. Same vendor, different contract.
Can Versely’s Omni Video row extend to 40 seconds?
No. Gemini Omni Video publishes 4s, 6s, 8s, and 10s. Forty seconds is the 1.1 cumulative cap on Google’s surface. For a native long take on Versely, use Seedance 2.5 (4–30s, 720p, from 34 credits).
Does more memory mean I can change the story at 0:20?
You can prompt a new direction. Ten seconds of context will still try to keep identity and space. A full recast is a new generate. Use extend to continue; use a fresh Omni Video or Veo call to relocate.