Gemini Omni 1.2 Flash: Price per Second, 15-Second Clips and What an Ad Really Costs

Google made Gemini Omni 1.2 Flash generally available on 5 November. It is the video model behind Google's Flow, Vids and Gemini app, and it is now on the paid Gemini API too. Clips run up to 15 seconds instead of 10, you can fix a clip with follow-up instructions instead of starting over, and Google says it costs 30% less: about $0.07 per second of 720p video. Four days before 11.11 and the Black Friday rush, that puts a 15-second product clip at roughly a dollar. This article covers what Google released, the full price table by resolution, what a real ad costs once you count the drafts, the limits that matter for a business in Pakistan or the Gulf, and what we would test before trusting it.

What Google released on 5 November

The facts below come from Google's launch post, its Cloud announcement and the Gemini API documentation, all read on 7 November.

• What it is. A video model that generates and edits clips with built-in sound. The model ID is gemini-omni-1.2-flash. It is generally available on the Gemini API and Google Cloud, and Google lists it in Flow, Google Vids, Google Ads and the Gemini app, where Google AI Ultra subscribers get it first.
• Longer clips. Native clips of 3 to 15 seconds, set in one-second steps. Omni 1.1 stopped at 10.
• Follow-up edits. Google says you can swap a product, change the lighting or the aspect ratio, and fix a mistake turn by turn without regenerating the whole clip.
• Drafts, then upscale. Make cheap drafts at 360p or 720p, then upscale the keeper to 1080p or 4K.
• Reference images. You can supply character, product and style images, and Google says the model keeps them consistent from the first frame to the last.
• Speech. Google claims more natural dialogue, clearer lip-sync and distinct voices for more than one speaker.
• Watermark. Every output carries an invisible SynthID watermark that tools can detect.

One thing the launch post does not contain is a benchmark score. Its comparisons with Omni 1.1 are example clips, and the customer quotes (invideo, Photoroom, Picsart, Runway and others) are Google's selection. We found no independent test of 1.2 yet. Treat every quality claim here as Google's.

The price table: $0.07 a second at 720p

Google bills video by output tokens: $12.00 per million for video, against $17.50 for Omni 1.1. At 720p one second is 5,792 tokens. These are the per-second figures on Google's pricing page, read on 7 November. There is no free tier for this model.

The 10 and 15 second columns are our arithmetic from Google's per-second rates, apart from the 720p figures, which Google prints itself. For comparison, Omni 1.1 was about $0.10 per second at 720p. So the cut is real, 31% on the token price, and it lands on the resolution most people actually use.

A cheaper second of video does not make a cheaper ad. It makes the first draft cheap. The retakes are still the bill.

What a 30-second ad really costs

Nobody publishes the first clip the model makes. Here is the arithmetic for a 30-second product ad, two 15-second clips, using Google's rates and one assumption we have to state: that an upscaled clip is billed at its own resolution's rate and the draft you upscaled from is billed separately. Google's table lists each resolution as its own line and does not say how a draft that is later upscaled is counted, so check your first invoice.

Now add the retakes. If you make five 360p drafts of the whole ad and one 1080p final, that is five times $0.70 plus $3.13, about $6.60 for a finished 30-second ad. Even if you throw away ten 720p attempts first, you are near $24. Compared with a studio shoot that is nothing, and you should say so to anyone still quoting you for stock footage. Compared with the old Omni it is about 30% cheaper, which matters if you make forty variants for forty products.

• The savings come from drafting small. Always iterate at 360p or 720p. Going straight to 4K costs three times more per second than 720p.
• The follow-up edit is the new money-saver. If a fix to one clip works, you pay for one clip, not the whole ad.
• Set a cap before you start. A script that loops on a failed prompt will spend quietly. Put a daily budget on the key.

What changed from Omni 1.1, in one table

Google's own model pages list both versions side by side. The resolutions (360p, 720p, 1080p, 4K at 24 frames per second) and the token limits are the same.

Google will route all gemini-omni-1.1-flash requests to 1.2 on 3 December and bill them at 1.2 rates. If you have an app on 1.1, the price drop arrives by itself, and so does any change in how your prompts behave. Test before that date, not after.

The limits that matter if you sell in Pakistan or the Gulf

The launch post is about what the model can do. The documentation is where Google lists what it cannot, and for our readers three lines matter.

• English is the supported language. The documentation says English is fully supported and that other languages have not been evaluated, so results may vary. If your ad needs spoken Urdu or Arabic, plan to record or dub the voice separately and use the model for the pictures. Do not assume a lip-sync claim that was demonstrated in English holds in another language.
• Reference limits are tight. Up to six reference images per request, as three images each of two characters. A 15-second clip accepts one reference video of up to three seconds; shorter clips accept up to three. Reference videos work only when generating, not when editing or extending.
• People are restricted. Uploading and editing images of certain recognizable people is not supported. Editing or extending uploaded videos is unavailable in the EU, Switzerland and the UK, though model-generated videos can be edited.
• Missing controls. No negative prompts, no system instructions, no voice editing. Audio references are gated behind an allowlist, not open. Google also says it is still working on fast sports action, rapid script changes and complex physics.

Where this leaves Veo

Google's video documentation now says the Veo 3.1 preview models in the Gemini API were shut down on 22 October, and recommends moving to Gemini Omni 1.2 Flash or the generally available models on Google's enterprise platform. We could not read a current Veo price from Google's pages, so we are not comparing prices with it. What the page does tell you is that for a developer on the Gemini API, Omni is now the video model Google points to. If you built something on a Veo 3.1 preview ID, it has already stopped working.

What we would test before trusting it

We build video tooling, and our AI video studio works the way a careful editor does: approve the look as a still first, make every shot match that locked frame, and re-render one shot rather than the whole film. That workflow exists because video models drift. Our own tests of the earlier Omni on 23 October found two things we had to design around. Asked to move from one approved still to another, the model did not accept a last frame and redrew the first one (the match to the supplied still was poor). And without an explicit sentence saying "one continuous take, no cuts", it cut three times in eight seconds.

We have not yet re-run those tests on 1.2, and Google's pages now describe first and last frame interpolation for clips of 3 to 10 seconds and fewer jump cuts. Those are exactly the claims to check first. Our list:

• Product fidelity. Give it your real product photos and compare the logo, label text and colour on the third clip, not the first.
• Continuity across clips. A 30-second ad is two 15-second clips. Check that the person, the room and the voice survive the join.
• Text on screen. Prices and phone numbers should be added in an editor afterwards. Do not trust any video model with digits you need to be right.
• Your language. Run the same prompt in English and in Urdu or Arabic and listen.
• The real cost. Count every attempt for one finished ad, and divide. That, not $0.07, is your price per ad.

Your next step: make one ad, count the attempts

The honest summary is that Google made a decent video model cheaper and longer in the same week the older Veo previews were switched off. That is good news for a shop that wants a product clip for 11.11 and does not have a studio. It is not yet proof that the output is good enough to put your brand on, and the only way to learn that is to make one ad and keep the receipt.

If you want the cost side for other AI work, see what a Gemini Live avatar video agent costs, the cheapest AI model by cost per task, and what AI costs a small business per month. If you would rather have a finished film than a prompt, see how our AI video studio works or talk to us about what you want to make.