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

Google shipped Gemini Omni 1.2 Flash on November 5, 2026, and if you make short-form video for a living, this is the update that changes your cost math. The new model generates and edits native video clips up to 15 seconds long — up from 10 seconds — and it costs $0.07 per second for 720p video, a 30% cut from the previous version.

That’s the headline. The part that matters for your wallet is what 15 seconds actually buys you: a full product shot, a complete ad beat, or a whole Reel scene in a single clip, without stitching two generations together and hoping the character’s face survives the cut. A full 15-second clip costs about $1.05. A 30-clip campaign costs roughly $31.50. And if you use Google Vids, the same clips cost you nothing extra at all.

This isn’t a launch announcement rewritten. Below is the costed creator workflow: what 15 seconds unlocks that 10 couldn’t, the real per-clip and per-campaign math, a step-by-step walkthrough for making one product ad in Flow or Vids, and the honest limits Google admits to — because the model still struggles with fast sports, quick dialogue changes, and complex physics.

What Gemini Omni 1.2 Flash actually is

Gemini Omni 1.2 Flash is Google DeepMind’s fast, cheap video generation and editing model. It sits in the Gemini family as the volume tool: not the flagship quality model, but the one built for businesses and creators who make videos in bulk — product demos, virtual try-ons, ads, and social clips.

The November 5 release, announced by Google DeepMind product manager Anish Nangia, makes the model generally available across the Gemini API and Gemini for business. General availability matters here because it means the model is now a production service with service-level agreements, not a preview you build a workflow around and then watch disappear. Design platforms are already building it in: Artlist, FLORA, Krea, and invideo all run it in production, according to Google’s own announcement.

Three things changed in this release:

• Longer clips. Native generation and editing up to 15 seconds, up from 10 seconds with Omni 1.1. You pick the duration in one-second steps from 3 to 15 seconds, and 10 seconds is the default.
• Lower price. $0.07 per second for 720p video, 30% less than Omni 1.1. Google didn’t print the old number in the announcement, but the arithmetic points to about $0.10 per second before — treat that as arithmetic, not a quoted price.
• Better consistency. Multi-turn conversational editing and stronger reference following. You can refine a clip by chatting with the model — swap a background, change the weather, adjust the aspect ratio — instead of regenerating from scratch and losing the version you liked.

The model generates synchronized video and native audio from text, image, and video inputs. That last part is the real shift for editors: you can hand it an existing clip or a product photo and ask it to transform it, rather than starting from a text prompt every time.

The cost math, worked out properly

Google’s headline rate is $0.07 per second of 720p video. On the paid Gemini API the meter is actually token-based — $12 per million video-output tokens, at 5,792 tokens per second of 720p — which works out to about $0.70 for a 10-second clip and about $1.04 for a 15-second clip. The per-second figure and the token figure agree; use whichever is easier to budget with.

Here’s what that means in practice:

And at campaign scale:

Two things worth noticing. First, the jump from 10 to 15 seconds is a 50% longer clip for 50% more money — there’s no bulk discount hidden in the longer format, but there’s no penalty either. Second, the draft-and-upscale workflow Google recommends is where the real savings live: draft at 360p or 720p, iterate cheaply, and upscale only the winning take to 1080p or 4K. Every failed draft at 720p costs you pennies; every failed draft at 4K would cost real money.

The migration date matters if you’re already using the API. Omni 1.1 stays available during a four-week transition, and on December 3, 2026, API requests to 1.1 automatically route to 1.2 with no code changes — and at the lower 1.2 rates. If you have a pipeline calling 1.1, you don’t need to touch it, but you should know the output will change on December 3.

What 15 seconds actually unlocks for a Reel or Short

Five extra seconds sounds small. In short-form video, it’s the difference between a scene and a fragment.

A 10-second clip gives you a hook and a product shot, and then it ends right as the viewer is deciding whether to keep watching. A 15-second clip gives you a hook, a product shot, and a payoff — the hand opening the box, the before-and-after, the punchline. For a product ad, that’s the difference between “here’s the thing” and “here’s the thing, here’s what it does, here’s why you want it.”

It also changes the editing math. A 30-second Reel built from 10-second clips needs three generations, and every seam is a place where the character, the lighting, or the product can drift. Two 15-second clips halve the seams. Google’s own numbers back this up: in DeepMind’s side-by-side evaluations, Omni 1.2 won 62% of visual-consistency comparisons against Omni 1.1 in text-to-video testing. That’s Google’s own evaluation, not an independent benchmark — but it’s the number that matters for anyone stitching clips together, because consistency across cuts is exactly what breaks AI video.

The other unlock is clip extension. The model can now extend a scene — take a finished clip and keep it going — which means a 15-second base clip can become a 30-second scene without starting over. Google demonstrated this with everyday footage earlier this year, extending a poolside clip into a longer continuous scene. For creators, that’s the closest thing to a timeline in AI video: generate the opening, extend it, then edit the result in conversation.

The workflow: one product ad, start to finish

Here’s how a 15-second product ad actually gets made with Omni 1.2 Flash, in the two places most creators will touch it.

Option 1: Flow (the filmmaker’s route)

Flow is Google’s AI filmmaking tool, and it’s where the model’s editing strengths show up best. The workflow looks like this:

1. Start with a reference. Upload a product photo — the actual product, not a stock image. Omni 1.2’s reference following is the improvement Google is proudest of, and it’s what keeps your product looking like your product instead of a generic version of it.
2. Prompt the scene in one sentence. Something like: “A ceramic coffee mug on a sunlit kitchen counter, steam rising, camera slowly pushes in, warm morning light.” Keep the camera move simple; the model handles slow pushes and static shots far better than fast pans.
3. Set the duration to 15 seconds. The default is 10 — change it. You’re paying $1.05 either way for a full-length clip, and the extra five seconds is where the payoff lives.
4. Edit in conversation. The first version will have something wrong with it — the steam will look like smoke, or the light will be too orange. Instead of regenerating, tell it: “Make the steam thinner and the light cooler.” Multi-turn editing is the feature that saves you the most money, because every regeneration is a fresh $1.05 and every edit is a refinement of the clip you already have.
5. Upscale only the winner. Once the 720p draft is right, upscale to 1080p or 4K for the final export. Google recommends exactly this order, and it’s the cheapest way to work.

Option 2: Google Vids (the free route)

If you have a Google account or a Workspace subscription, 15-second Omni clips in Google Vids cost nothing extra. That’s the route most small creators should try first. Vids is built for business video — the same product demos and ads the model is tuned for — and the clips you generate there drop straight into a timeline you can add text, voiceover, and branding to.

The trade-off is control. Vids is a presentation-style editor, not a filmmaker’s tool, so you get less fine-grained control over camera moves and reference handling than Flow gives you. For a straightforward product ad or a social clip, that’s usually enough. For anything where the shot needs to look a specific way, Flow is worth the per-clip cost.

The campaign version

Say you’re running a 30-clip campaign — one product, 30 variations for different audiences or ad sets. The math is $31.50 at 720p for the full set, before you count the drafts that didn’t make it. Realistically, budget for two to three drafts per keeper when you’re starting out, which puts a 30-clip campaign closer to $60–$95 in generation costs. That’s still cheaper than a single hour of a videographer’s time, and it’s the number to put in front of a client who asks why AI video is worth trying.

Omni 1.2 Flash vs Veo 3.1: which one do you actually need?

Google now sells two video models that look similar from the outside, and picking the wrong one wastes money. Here’s the honest split.

Veo 3.1 is the quality flagship. If you need the most cinematic output, the most realistic motion, and the best audio sync, Veo is the model — and it costs accordingly. We covered Veo 3.1 in depth in our guide to making Reels with Veo 3.1, and the verdict there still holds: Veo is for the hero clip, the one video that has to look expensive.

Omni 1.2 Flash is the volume model. It’s faster, it’s cheaper per second, and — critically — it’s an editing model, not just a generation model. Veo makes a clip; Omni makes a clip and then lets you change the background, extend the scene, and swap the aspect ratio in conversation. For a creator making 30 variations of one ad, or a business making product videos in bulk, that’s the difference between a tool and a toy.

The rule of thumb: if you’re making one video that has to be perfect, use Veo. If you’re making thirty videos that have to be good and consistent, use Omni. And if you’re making videos inside Google Vids, you’re using Omni whether you think about it or not — which is fine, because it’s free there.

What creators are actually saying

The early reaction from people who build with these tools has been cautiously positive, and the praise has a specific shape: it’s about reliability, not magic.

Shivam Rastogi, VP of Engineering at invideo, put it this way in Google’s announcement: “We ran Gemini Omni 1.2 Flash through our evals at invideo. Honestly, it’s really good. It sticks to the prompt, and the physics and sound feel real. If you need fast video at scale, it’s one of the best we’ve tested. And 15 seconds means a full scene in one go.”

That’s a vendor quote in Google’s own blog, so read it with that in mind — but the substance of it matches what the model’s design promises. The “sticks to the prompt” part is the one creators care about most, because prompt drift is the tax every AI video tool charges. A model that follows a reference image closely and keeps a character consistent across a 15-second clip is worth more than a model that makes prettier 8-second clips.

The skepticism is real too. In Reddit threads comparing AI video tools, creators have consistently complained about prompt adherence — models adding random elements, drifting from the requested style, or producing smooth motion when the prompt asked for stop-motion. Omni 1.1 drew exactly that criticism. Omni 1.2’s answer is the 62% consistency win rate and the reference-following improvements, but those are Google’s numbers, and no independent benchmark has verified them yet. The honest position: try it with your own product photo and your own prompt before you commit a campaign budget to it. The test costs about a dollar.

The limits Google admits to — and the ones it doesn’t

Google is unusually upfront about one thing: the model still struggles with fast sports, quick script changes, and complex physics. If your ad involves a basketball dunk, a rapid costume change, or a Rube Goldberg machine, expect artifacts. Slow, deliberate scenes — a product on a table, a person walking through a room, a landscape at golden hour — are where it shines.

The limits Google doesn’t advertise are the ones you’ll find in the first hour of use:

• Faces drift over long clips. Fifteen seconds is long enough for a face to subtly change. Reference images help, but they’re not a guarantee. If your ad has a human spokesperson, test the face across the full 15 seconds before you build a campaign around it.
• Text in the video is unreliable. AI video models still garble on-screen text. Don’t ask the model to render your brand name or a price — add text in your editor afterward.
• The Gemini app rollout is gradual. Fifteen-second videos in the Gemini app are rolling out to Google AI Ultra subscribers first. If you’re not on Ultra, or the rollout hasn’t reached you yet, the app won’t show the feature. The API, Flow, and Vids are the reliable access points right now.

SynthID, watermarks, and the disclosure question

Every clip Omni 1.2 Flash produces carries an invisible SynthID watermark. You can’t see it, but it’s detectable as a marker of AI origin. That’s a feature, not a bug — it protects you when someone asks whether your video is real, and it protects your audience from being misled.

But the watermark doesn’t replace disclosure. If you’re posting AI-generated video to Instagram, Facebook, TikTok, or YouTube, those platforms have their own AI-label rules, and a SynthID watermark doesn’t automatically satisfy them. Label your AI content as AI content. It’s also just good business: the audience that discovers your video was AI-generated after the fact is an audience that stops trusting you, and the audience that knows upfront judges the video on its merits.

One more practical note: the watermark survives the upscale to 1080p or 4K, and it survives editing in most tools. Don’t count on cropping or re-encoding to remove it — and don’t try. The point of the watermark is provenance, and stripping it is the kind of thing that gets a creator in trouble with a platform.

Where to use it, and what it costs you there

Here’s the access map as of November 6, 2026:

If you’re a solo creator, start with Google Vids — it’s free, and it tells you whether the model’s output is good enough for your use case before you spend anything. If you’re a business making product videos at volume, the API or Flow is the serious route, and the $31.50-per-30-clips math is the number to budget around. If you’re already calling Omni 1.1 in a pipeline, mark December 3 on your calendar: that’s when your requests start routing to 1.2 automatically.

For the Vids workflow specifically, our guide to making free AI video with Google Vids and Gemini walks through the editor step by step — the Omni 1.2 clips slot into exactly that workflow, just longer and cheaper than before.

The bottom line for creators

Gemini Omni 1.2 Flash isn’t a revolution. It’s a 50% longer clip, a 30% cheaper second, and an editing model that finally lets you fix a video instead of starting over. Those are incremental improvements, but they land exactly where creators feel the pain: the clip that ends too soon, the bill that adds up across a campaign, and the regeneration loop that eats an afternoon.

The test is cheap. Make one 15-second product clip in Google Vids for free, or spend $1.05 on the API. If the product looks like your product and the scene holds together for the full 15 seconds, you have a volume video tool. If the face drifts or the physics fall apart, you know before you’ve spent real money — and you know Veo is the better call for that particular shot.

Either way, the December 3 auto-migration means the decision is partly made for you. If you’re already in the Gemini video ecosystem, 1.2 is coming. The only question is whether you use the next four weeks to learn the new clip length and the new editing workflow, or spend them making 10-second videos that cost 30% more.

Frequently Asked Questions

How much does a 15-second Gemini Omni 1.2 Flash video cost?

About $1.05 at 720p through the Gemini API, at Google’s published rate of $0.07 per second. On the API’s token meter that’s roughly $1.04 per 15-second clip. In Google Vids, 15-second clips cost nothing extra with a Google account or Workspace subscription.

Is Gemini Omni 1.2 Flash free anywhere?

Yes — in Google Vids, where the clips are included at no extra cost with a Google account or Workspace subscription. Everywhere else (Gemini API, AI Studio, Google Cloud, Flow, Google Ads) you pay the per-second rate. The Gemini app rollout to AI Ultra subscribers includes it in the subscription, but that rollout is gradual.

What’s the difference between Omni 1.2 Flash and Omni 1.1?

Three things: clips run up to 15 seconds instead of 10, the price dropped 30% to $0.07 per second at 720p, and the model got multi-turn conversational editing plus stronger reference following. Omni 1.1 requests automatically route to 1.2 on December 3, 2026.

Can Gemini Omni 1.2 Flash make vertical video for Reels and Shorts?

Yes. You can set the aspect ratio as part of your prompt or edit it in a follow-up turn — the multi-turn editing handles aspect-ratio swaps. A 15-second vertical clip is a natural fit for a Reel or Short, though you’ll still want to add captions and on-screen text in your editor rather than asking the model to render text.

Do Omni 1.2 Flash videos have a watermark?

Every clip carries an invisible SynthID watermark that marks it as AI-generated. It’s not visible to viewers, but it’s detectable. You should still label AI-generated content as AI on Instagram, Facebook, TikTok, and YouTube, since the watermark doesn’t replace platform disclosure rules.

When should I use Omni 1.2 Flash instead of Veo 3.1?

Use Omni when you’re making video at volume — product demos, ad variations, social clips — and when you need to edit a clip in conversation rather than regenerate it. Use Veo 3.1 when you’re making one hero clip where maximum cinematic quality matters most. If you’re unsure, test the same prompt in both: the $1.05 Omni test is cheap, and the comparison tells you more than any review will.