Gemini Omni 1.2 Flash: 15-Second AI Video for $0.07 a Second

Google DeepMind shipped Gemini Omni 1.2 Flash on 5 November 2026, and it is generally available from today across the Gemini API and Gemini for business. The headline is simple enough to fit in one sentence: native 15-second video generation and editing, with a 30% price cut to $0.07 per second for 720p video.

I have not tested it hands-on, and this article does not pretend otherwise. Everything below comes from Google's own announcement, checked today. What I can do is the part that matters more than the press release: work out what the numbers actually mean for your wallet, your workflow, and the clips you would realistically make.

What changed vs Omni 1.1

The 15-second number deserves a second look, because it is the change that unlocks the rest. Ten seconds is a teaser. Fifteen seconds is a complete product explainer, a full social ad spot, or a Short with a beginning, a middle and an end — Google says exactly that in the announcement, and for once the marketing line holds up. You can also pick the length in one-second increments from 3 to 15 seconds, which sounds minor until you have paid per-second pricing for a 10-second clip when you only needed six.

The price cut, in real money

Thirty percent off sounds abstract. Here is what it costs in practice, at $0.07 per second for 720p:

• A 5-second clip: $0.35
• A 10-second clip: $0.70
• A full 15-second clip: $1.05

A hundred 15-second clips costs $105. That is cheap enough that the generation itself stops being the budget question — the question becomes how many takes you burn getting one you like. Google's answer to that is built into the workflow: draft fast at 360p or 720p, fix details with a follow-up edit instead of regenerating the whole clip, then upscale your best creations to 1080p or 4K.

That workflow is the real price story. Per-second pricing punishes re-rolls, and every video model lives or dies on re-rolls. If conversational editing works the way Google describes it — swap a product or background, change the lighting or weather, switch the aspect ratio, all on the same clip — then the savings are not really the 30%. They are the regenerations you never pay for. If it does not work that way in practice, you will find out on your second or third clip, and the bill will tell you first.

One honest note on the old price: Google announced a 30% cut to $0.07, which implies Omni 1.1 sat at $0.10 per second. I am stating that as arithmetic, not as a quoted price — Google did not print the old number in the announcement.

Conversational editing is the feature to watch

Most video models treat every change as a fresh generation. You liked the shot but the jacket is wrong? Start over, pay again, hope the new version keeps everything else. Omni 1.2 lets you refine the same clip turn by turn instead.

Google's examples are practical rather than flashy: swap a product or a background, change the lighting or the weather, switch the aspect ratio. The use case they keep coming back to is ad variants — one clip, re-cut for different audiences and placements without re-briefing or re-generating. Their demo runs a single generated clip through several backgrounds in sequence — a striped pavilion tent, an oak forest, a stable with a white horse, a market fair, a stone castle — while keeping both characters' performances, armor reflections and camera framing intact.

That last clause is doing a lot of work. "Keeping the performance intact" is the whole promise, and it is also the part no announcement can prove. Background swaps that quietly change the actors' faces are the classic failure mode of this kind of editing. Until someone runs it on real product shots, treat it as a claim with a good demo, not a settled fact.

Reference following: the try-on test

The other big claim is reference adherence. You can guide generation with character, product or style images, short motion clips up to three seconds, and audio cues — and the model is supposed to hold onto those details from the first frame to the last. Faces, garments, packaging typography and brand colors stay consistent across camera angles.

Google's sharpest example is a virtual try-on: one model photo plus front-and-back product shots of three outfits, turned into a 15-second sequence with 360-degree turns and match cuts, keeping the model's identity and each garment's cut, drape and back details consistent. If that holds up on real catalog photography, it is genuinely useful — turning existing product photos into video ads without a shoot. If the garment drifts between angles, it is a demo, not a workflow.

The same logic applies to brand work. Packaging typography and brand colors staying consistent across a clip is the difference between an ad you can ship and a clip you have to explain away. Google's poster example — building a studio exhibition poster layer by layer, locking every material, shadow and typographic detail through the final frame — is the same claim from another angle. Consistency is the theme of this release, and consistency is also the thing that is hardest to verify from a launch post.

More natural speech, steadier cameras

Two more improvements are worth a paragraph each, because they target the two things that make AI video feel fake fastest: voices and camera moves.

On speech, Omni 1.2 promises more fluid, natural-sounding dialogue, clearer lip-sync and distinct multi-speaker conditioning — each sampled voice routed to the right character. Google's example is a 15-second animated scene with four character references and one-second voice samples, with expressive lip-sync and comedic timing. This shows up in Flow, in Gemini for business explainers, and in Google Vids with avatars, music and narration.

On cameras, the model is supposed to direct complex, continuous movements — sweeping FPV flythroughs, dramatic zoom-outs, orbits — with significantly fewer unwanted jump cuts, steadier scale and a more polished, cinematic look. The demo is a single continuous FPV drone flight across a piazza, through bronze doors, and up through the Pantheon's dome to the oculus, with zero jump cuts. A single continuous shot is the hardest thing for a video model to hold together, so if that demo is representative, it matters. Demos, of course, are chosen because they worked.

Where you can actually use it today

This is a general-availability launch, not a waitlist, and the access points are spread across Google's products:

• Developers: Gemini API, prototype in Google AI Studio, with developer docs and pricing.
• Businesses: Google Cloud console, enterprise solutions through the API, and Provisioned Throughput for scale. Outputs carry an imperceptible SynthID watermark, and the launch is backed by Google Cloud infrastructure and security.
• Everyone else: direct scenes and multi-turn edits in Flow; 15-second clips at no additional cost in Google Vids with a Google account or Workspace subscription; 15-second videos in the Gemini app, rolling out to Google AI Ultra subscribers; and scaled video campaigns in Google Ads.

The Google Vids detail is easy to miss and worth flagging: if you already use Vids, the longer clips cost nothing extra. For a lot of small teams, that is where this release will actually be touched first — not the API, not Flow, but the slides-adjacent tool they already open.

Who this matters for

• Creators making Shorts, Reels and ads. Fifteen seconds with dialogue, a beginning and an end, at about a dollar a clip, is a real format — not a teaser. The one-second duration steps mean you stop paying for seconds you do not need.
• Developers building video features. The API is GA, the pricing is per-second and predictable, and the edit-instead-of-regenerate workflow is the thing to test first. Draft cheap, upscale the winners.
• Brands and ecommerce teams. Virtual try-on from existing product photography and consistent packaging typography are the two claims that would change a real budget if they hold. Test them on your own catalog before believing the demo.
• Small teams already on Google Vids. Fifteen-second clips at no additional cost is a free upgrade to a tool you already have. Try it there before spending anything on the API.
• Anyone tracking the video-model race. Longer clips, conversational editing and reference consistency are where every video model is heading. This release is Google pulling those threads together in one version, with a price cut attached.

Should you switch today? A quick checklist

• You are paying per re-roll on another video model. Try the conversational edit flow on one real clip. If it holds the performance across edits, the savings compound fast.
• You make product or ad videos from existing photos. Run the try-on or catalog test on your own images. Do not judge it on Google's model and outfits.
• You are already in Google Vids or Flow. This costs you nothing new to try. Start there.
• You need long-form video. Wait. Fifteen seconds is still a cap, and Google says so itself.
• You need fast sports action, rapid script changes or complex physics. Google lists these as areas still being improved. Believe them.

The limits — stated plainly

• Fifteen seconds is still a cap. Longer than ten, still short. Anything with a real narrative arc needs clips stitched together, and Google's own advice is to extend clips and interpolate between keyframes.
• Google itself lists the weak spots. Fast sports action, rapid script changes and complex physics are, in Google's words, still being improved. That is unusually candid for a launch post, and it is the most useful sentence in it.
• No public benchmarks. Google published no benchmark numbers, no comparison table against other video models, and no independent evaluation. Every quality claim in this article is Google's claim, labelled as such.
• Consistency is claimed, not yet measured independently. Reference following, lip-sync and camera continuity all rest on Google's demos. Hands-on testing by anyone outside Google has not happened yet at the time of writing.
• The Gemini app rollout is gradual. Fifteen-second videos there are rolling out to Google AI Ultra subscribers — if you are not on Ultra, or the rollout has not reached you, the app will not show it yet.
• The old price is arithmetic, not a quote. Thirty percent off $0.07 implies about $0.10 per second for Omni 1.1. Google did not print the old figure.
• No hands-on testing was performed for this article. Nothing here is a hands-on finding, and no prices, dates, availability details or benchmark numbers beyond Google's announcement have been invented.

The bottom line is a practical one. Gemini Omni 1.2 Flash is a genuinely useful release, and its most useful part is not the headline number. Fifteen seconds matters, and $0.07 a second matters, but the thing that changes the economics is editing a clip instead of regenerating it — if it works on your footage the way it works in Google's demo. Try it where it is free first, in Vids or Flow. Then run one real product shot through the reference flow, one real clip through three conversational edits, and count the re-rolls you did not pay for. That count is the review.

Have you tried the conversational editing yet — and did it actually keep your subject intact across edits, or did the faces drift? That is the test I would trust more than any launch demo, and I would like to hear how it went.

Sources: Google — "Gemini Omni 1.2 Flash: 15-second video and lower pricing", 5 November 2026, https://blog.google/innovation-and-ai/technology/developers-tools/introducing-gemini-omni-1-2-flash/. All facts checked 5 November 2026. No hands-on testing was performed for this article. This breaking article is not the daily buyer-intent post, and it contains no affiliate links.