Gemini Omni 1.2 Flash: Google's 15-Second AI Video Model — What It Does, What It Costs, and the 4K Catch

Gemini Omni 1.2 Flash at a glance

What is new versus Omni 1.1

Omni 1.1 Flash was the preview version. Version 1.2 is the first generally available release of the Gemini Omni family, and the changes are practical rather than cosmetic: longer clips, cheaper seconds, and edits that do not throw away the clip you already have. Omni 1.1 stays available during a four-week transition — on 3 December 2026, API requests to 1.1 will automatically route to 1.2 at the lower price, with no code changes needed.

The price note needs one honest line: Google announced the 30% cut to $0.07 but did not print the old price, so the $0.10 figure for 1.1 is arithmetic from the percentage, not a quoted price. The $0.07 rate itself is Google's published number.

What it actually does better

Google's announcement leads with five improvements, each shown with its own demo clip. Here they are, in plain words:

1. Longer clips up to 15 seconds. Duration is settable in one-second steps from 3 to 15 seconds. That is the change that unlocks the rest: a complete product explainer, a full social ad spot, or a Short with a beginning, a middle and an end — in a single generation instead of stitching short clips together.

2. Better reference following. You can guide generation with character, product or style images, short motion clips of up to three seconds, and audio cues — and the model holds onto those details from start to finish. Faces, garments, packaging typography and brand colours stay consistent across camera angles. Google's flagship example: one model photo plus front-and-back product shots of three outfits becomes a 15-second virtual try-on sequence with 360-degree turns and match cuts, keeping the model's identity and each garment's cut, drape and back details consistent. Watch that exact clip in the sample below.

3. More consistent results. Stronger character, object and scene consistency from the first to the last frame, with less visual drift or morphing across motion and camera transitions. That also covers interpolating between a first and last keyframe, and extending a clip into a longer sequence. In one demo, the 1.1 model fails to match a static studio-lighting exhibition poster as closely as 1.2 does — version 1.2 builds the poster layer by layer and locks every material, shadow and typographic detail through the final frame.

4. Editing in conversation. Refine the same clip turn by turn — swap a product or background, change the lighting or weather, or switch the aspect ratio — instead of starting over. Google's demo starts from one generated clip and swaps the background turn by turn, from a striped pavilion tent to an oak forest, a stable with a white horse, a market fair and a stone castle, while keeping both characters' performances, armour reflections and camera framing intact. For ad teams, that means variants for different audiences and placements without re-briefing or re-generating.

5. More natural speech and camera work. Dialogue sounds more fluid, lip-sync is clearer, and different speakers get distinct voices — Google's demo routes each sampled voice to the right character across a 15-second animated conservatory scene with four characters. Camera direction improves too: continuous moves like sweeping FPV flythroughs, dramatic zoom-outs and orbits, with fewer unwanted jump cuts. One demo flies a single continuous FPV drone shot across a piazza, through bronze doors and up through the Pantheon's coffered dome to the oculus, with zero jump cuts.

Sample: watch what it makes

The clips below are Google's own official demo clips for this launch — the virtual try-on sequence and the conversational background-swap demo. They are Google's examples, not independent test results, and they show the model at its best.

Sample: Gemini Omni 1.2 Flash virtual try-on demo — one model photo plus three outfits becomes a 15-second try-on sequence. Official Google demo clip.

More official samples: Google's announcement page carries the full set of demo clips — the conservatory dialogue scene, the Pantheon FPV flight, the background-swap sequence and the poster build. The background-swap and conversational-editing demos are also in the article's files as reference.

A 15-second video of a street-food vendor in Lahore flipping parathas on a smoking griddle, one continuous slow push-in, no cuts, warm and energetic, with natural ambient sound and a line of dialogue.

How to use Gemini Omni 1.2 Flash

In Google AI Studio (fastest for developers). Open AI Studio, go to video generation, and select Gemini Omni 1.2 Flash. Type a prompt, or add reference images first — a product photo, a character sheet, a style image — then set the duration anywhere from 3 to 15 seconds and generate. Draft at 360p or 720p, then upscale the keepers to 1080p or 4K.

In Flow. Flow is Google's filmmaking tool, and Omni 1.2 Flash is available there for scenes and multi-turn edits — direct a shot, then keep editing the same clip in conversation instead of regenerating it.

In the Gemini app. Create 15-second videos in the Gemini app, rolling out to Google AI Ultra subscribers. This is the simplest route if you only want a clip, not an API workflow.

In Google Vids. Generate 15-second clips at no additional cost with a Google account or Google Workspace subscription, for explainers and internal videos with avatars, music and narration.

For businesses and ads. Use the Google Cloud console and the Gemini API for production workloads, with Provisioned Throughput available for steady volume. Google Ads can scale video campaigns on the model. Photoroom already uses it in its Video Ad Maker and Product Animator.

Honest limitations

• 720p is the native resolution. 4K comes from upscaling after generation. If your delivery must be native 4K from the first frame, Veo 3.1 is Google's native-4K route — with shorter 8-second generations.
• 15 seconds is the maximum per generation. Longer pieces come from extending or interpolating between clips, or joining clips in an editor. A 15-second cap is generous for an ad, but it is not a short film.
• Every clip carries a SynthID watermark. It is imperceptible to viewers but detectable as a marker of AI origin. Plan for that in client work — disclose AI generation.
• Reference following is better, not perfect. Google's own numbers are a 62% visual-consistency win rate against 1.1 in side-by-side tests — meaning 1.1 still won 38% of those comparisons. Check faces, hands, packaging text and logos frame by frame before delivery.
• Price is per generated second. A 15-second clip costs $1.05 at 720p, but retries, re-generations and variants add up. The calculator above prices one clip; budget for three or four attempts per finished shot.
• Availability is rolling out. The Gemini app route is still rolling out to AI Ultra subscribers, so not every account sees it on day one.

Security and privacy

• Your prompts and references go to Google. Text, images, audio and video references are processed by Google's services under its terms — do not upload unreleased products, private people, or client assets you have no right to share.
• Outputs are marked. Every clip carries an imperceptible SynthID watermark. Google positions this as provenance, not a restriction — but it means AI origin is detectable.
• Enterprise route exists. Gemini Omni is now generally available with Google Cloud's enterprise-grade infrastructure and security, and Provisioned Throughput is available for production workloads.
• People and likeness need care. The model follows faces and voices from references closely — get consent before generating a real person's likeness or voice, and never use it to impersonate.

Real use cases

• Product ads at scale: one product shoot becomes dozens of 15-second ad variants through conversational edits — different backgrounds, lighting and aspect ratios per audience and placement.
• Virtual try-on and catalog video: turn existing product photography into try-on clips and catalog spins without a reshoot, as in Google's demo.
• Social spots: 15 seconds is exactly a full ad spot or Short with a beginning, middle and end.
• Explainers in Google Vids: narrated videos with avatars, music and dialogue for training and internal comms.
• Film pre-visualisation: block a scene, test an FPV camera move, and check continuity before committing to a shoot.

Frequently asked questions

Is Gemini Omni 1.2 Flash really 4K?

Not natively. It generates at 720p (draft at 360p), and you upscale the best clips to 1080p or 4K afterwards. A "15-second 4K video" from this model is a 15-second native clip, upscaled to 4K. Veo 3.1 is Google's native-4K video model.

How long are the clips?

From 3 to 15 seconds, set in one-second steps, with 10 seconds as the default. Omni 1.1 topped out at 10 seconds.

How much does it cost?

$0.07 per second of 720p video on the Gemini API — $1.05 for a full 15-second clip, before retries. That is 30% less than Omni 1.1. Google Vids generates 15-second clips at no additional cost with a Google account or Workspace subscription.

Where can I use it?

Gemini API, Google AI Studio, Google Cloud console, Flow, Google Vids, the Gemini app (rolling out to AI Ultra subscribers) and Google Ads.

Can it edit a video I already made?

Yes — multi-turn conversational editing works on the same clip. Ask for a different background, lighting, weather or aspect ratio and it edits that clip instead of generating a new one.

What happened to Omni 1.1?

It stays available during a four-week transition. On 3 December 2026, API requests to 1.1 automatically route to 1.2 at the lower price, with no code changes.

Does it add a watermark?

Yes. Every output carries an imperceptible SynthID watermark marking it as AI-generated.

Sources

• Google — "Gemini Omni 1.2 Flash: 15-second video and lower pricing" (5 November 2026) — announcement, specs, pricing, demo clips
• Unite.AI — "Google's Gemini Omni 1.2 Flash Reaches General Availability" — token pricing ($12 per million video-output tokens), 1.1 transition date, 62% consistency figure
• The Decoder — launch coverage — demo clips referenced in the sample section
• LaoZhang AI Blog — Seedance 2.5 vs Veo 3.1 vs Gemini Omni — Veo 3.1 specs used in the comparison (8-second generations, native 4K, extension to 148 seconds)

Related reading: Teardown: the free Claude Code skill that turns any machine into a 3D simulator · Kimi K3.1 Pro: open weights, 2.8 trillion parameters · 10 GitHub repos to build your own overnight system