Gemini Omni 1.2 Flash Extend

Gemini Omni 1.2 Flash Extend: AI Video Extension

What is Gemini Omni 1.2 Flash Extend?

Gemini Omni 1.2 Flash Extend, from Google, continues a source video by 3 to 10 more seconds and returns the original clip followed by the new seconds as a single file. The source plus the added seconds can total up to 30 seconds, so a 27-second clip takes a 3-second extension and a 25-second clip the default 5. Part of the Gemini Omni family, this model generates picture and native synchronized audio together, so the added seconds carry their own sound rather than a silent tail. Leave the prompt empty to continue the scene naturally, or describe the next beat to steer what happens.

Key Features

• •Extend any source video by 3 to 10 added seconds; the output is the source plus those seconds, up to 30 seconds total.
• •Native synchronized audio generated across the added seconds, not a silent tail.
• •Optional prompt to direct the next beat, or an empty prompt for a natural continuation.
• •Output from 360p through 4K; leave resolution empty to keep the source's, or set one to re-render the whole clip.
• •A returned interaction id that Gemini Omni 1.2 Flash Refine can pick up within 24 hours for a follow-up.
• •SynthID watermarking and C2PA Content Credentials embedded on every output.

Best Use Cases

Gemini Omni 1.2 Flash Extend is strongest on stylised, no-people content where the join between the source and the continuation stays clean: carrying a motion-graphics explainer forward by one more step, holding a title card a few seconds longer, or stretching a stylised loop into a longer sequence. Because the output is billed for the added seconds only, extending a short clip is lighter work than regenerating a long one from scratch. In testing, a flat 2D vector "CITY RAINFALL" infographic was extended by 5 seconds while a dashed trend line drew across the bar tops into an up-arrow, every label staying legible and the style holding across the join, with matched native audio.

Prompt Tips and Output Quality

Describe only what happens next, in order (first..., then...), and name the stylised medium so the continuation matches the source. Write one sound per beat. Leave resolution empty to keep the source's resolution and keep on-screen text crisp; setting a resolution re-renders the whole clip, so long 4K extends run for minutes. Check the frames at the join and keep a take only where the continuation is visibly smooth. Stylised scenes without people hold up best; photoreal people are weaker and can trip content-safety filters.

How to Use Gemini Omni 1.2 Flash Extend

Gemini Omni 1.2 Flash Extend lengthens an existing clip. Upload a source video in video (27 seconds or less) and set duration to the number of seconds to add (3-10). The output is the source followed by the added seconds, up to 30 seconds total, so trim the source first if it is close to the limit.

Write the continuation prompt

Leave prompt empty for a natural continuation of the scene, or describe only what happens next. Give beats in order (first..., then...), name the stylised medium so the continuation matches the source, use a locked-off camera where the source has one, and list what must stay still. Write one sound per beat; native synchronized audio is generated for the added seconds.

Keep the join clean

Leave resolution empty to keep the source's resolution: the source frames are preserved and only the added seconds are generated, which keeps on-screen text crisp. Setting a resolution re-renders the whole clip, so long 4K extends run for minutes; use the asynchronous API for those and check the frames at the join before you rely on a take.

Recommended Settings

• •Carry an explainer or data video forward: leave resolution empty, describe one new beat and name the medium. Tested: a flat 2D vector "CITY RAINFALL" infographic, a 6-second source extended by 5 seconds, drew a dashed trend line across the bar tops into an up-arrow while every label stayed legible with matched native audio.
• •Natural continuation: leave prompt empty and the scene simply carries on from its final frames.
• •duration: 3-10 seconds added; keep source plus added within 30 seconds (at most 27s source with +3s, 25s with +5s, 20s with +10s).
• •seed: -1 for variation; pin an integer to reproduce a take.

Keep clips stylised and without photoreal people, which are weaker and can trip safety filters. Every run returns an interaction id that Gemini Omni 1.2 Flash Refine can continue within 24 hours on the same account.

Examples to try

Each example ran end to end on this model at the settings shown; every clip is the source followed by the added seconds, with the source's framing and native audio carried across the join.

• •

  Prompted next beat on a 2D infographic (default example). Prompt: "A dashed yellow trend line draws itself from left to right across the tops of the three bars and ends in a small arrowhead pointing up. Same flat 2D vector motion-graphics style, deep teal background, bold white sans-serif type, locked-off camera. The 'CITY RAINFALL' title, the three labelled bars and the 'MM PER MONTH' tag all stay exactly where they are. No people, no photorealism. Sound: a soft pen-stroke whoosh as the line draws, a light chime when the arrowhead lands. No new text." The dashed line draws into an up-arrow while every label holds; name the exact 2D style and list what stays still so text survives the added seconds.

• •

  Natural continuation with no prompt. Prompt: left empty. A clay stop-motion triceratops keeps waddling across its meadow past the join, look and sound carrying on. Leave the prompt empty when you want the source to simply run on, and feed a clip that ends mid-motion.

• •

  Native audio across a prompted beat sequence, in low-poly 3D. Prompt: "Same low-poly 3D style, single continuous shot, no scene cuts, locked-off camera. First a puff of grey smoke rises from the cabin's stone chimney, then the front window lights up yellow, then two low-poly birds cross the pale sky. The faceted hill and cabin stay put. Sound: a wood crackle as the smoke rises, a low chime as the window lights, two wing-flaps. No new text, no people." In the added seconds two low-poly birds cross the sky as the new beat, each sound landing on its action; name the medium and write one sound per beat.

Also built for continuations like "Continue the scene; the camera keeps panning left", "A clay figure walks in from the right and waves", or "Hold the shot; the candle flame flickers".

How long can the extended video be?

The source plus the added seconds, up to 30 seconds total.

Do I need a prompt to extend a video?

No. Leave it empty for a natural continuation, or add one to direct the next beat.

Does the added part have sound?

Yes, native synchronized audio is generated for the added seconds.

What resolutions are supported?

360p, 720p, 1080p and 4K; leave it empty to keep the source's resolution.

Can I keep extending or refining?

Pass the returned interaction id to Gemini Omni 1.2 Flash Refine within 24 hours on the same account.