Google Releases Gemini Omni 1.2 Flash: 15-Second Video at $0.07 per Second for 720p, a 30% Cut From Omni 1.1; Quality Rests on Demos and Customer Quotes, and Audio References Need Allowlisting

On November 5, 2026, Google announced Gemini Omni 1.2 Flash, a video generation and editing model, and said it is generally available across the Gemini API and Gemini for business. The announcement comes in two posts: a Google DeepMind product post and a Google Cloud post about production use. Every number below is Google's own, and the post gives no outside measurement of quality or speed. A captured Google Cloud post instead says Gemini for business chat is “coming soon”; API/Cloud availability and business-chat availability are separate, and we cannot resolve the two posts’ timing.

The main change is clip length. Omni 1.2 Flash can generate and edit video natively up to 15 seconds, up from 10 seconds in Omni 1.1, and the length can be set in one-second steps from 3 to 15 seconds. The Google Cloud model page adds that output from a first and last frame, or an extension of an earlier clip, is limited to 3 to 10 seconds, and that it lists 16:9 and 9:16 as the supported aspect ratios. Google says the model follows references more closely. A request can combine character, product or style images, short motion clips of up to three seconds, and audio cues, and the model is meant to hold faces, garments, packaging typography and brand colors steady across camera angles. Clips can also be refined turn by turn in a conversation, for example by swapping a background, changing the lighting or switching the aspect ratio, instead of generating the whole clip again. Google adds work on speech: more natural dialogue, clearer lip-sync and separate voice conditioning for several speakers.

Price is the one hard number in the post. Google says it cut the price by 30% from Omni 1.1, and that 720p video now costs $0.07 per second. By our arithmetic, a full 15-second clip at 720p is about $1.05 before retries. The post does not give prices for 1080p or 4K output, or for upscaling. It describes a cheaper workflow instead: draft at 360p or 720p, fix a detail with a follow-up edit, then upscale the best result to 1080p or 4K. It links to a pricing page, which we did not capture, so this article does not restate any other price.

Availability differs by product. Developers can use the model in Google AI Studio and the Gemini API. Businesses can use it in the Google Cloud console and through the API, where Google says Provisioned Throughput is offered and outputs carry an imperceptible SynthID watermark. Google also lists Flow, Google Vids, where 15-second clips are generated at no additional cost with a Google account or a Google Workspace subscription, and Google Ads. The Gemini app gets 15-second videos as a rollout to Google AI Ultra subscribers. The Google Cloud post adds a limit that the DeepMind post does not: audio references are gated behind allowlisting, so not every developer can use them on day one.

The evidence for quality is demonstration videos and customer quotes. Google shows a virtual try-on sequence, a poster built layer by layer, a five-step background swap and a single FPV flight through a domed building with no cuts, and in some of them compares Omni 1.2 with Omni 1.1 side by side. There is no benchmark score, no preference study, no latency figure and no failure rate. Quotes come from Adobe, Monks, Whering, Figma Weave, STAN, Magnific and Resemble AI in the DeepMind post, and from Artlist, FLORA, invideo, Krea, Picsart, Runway, Photoroom and QuickFrame AI in the Cloud post. They are chosen by the vendor. One of them, QuickFrame AI, says 15-second generation at 1080p lets users make a complete connected-TV ad in one generation.

Google also states what is not done yet. It says development continues on fast sports action, rapid script changes and complex physics. That is a short list, and it matches what a team would want to test first: motion with high speed, scenes that change quickly and objects that interact. The post does not describe any safety evaluation or content policy for the model beyond the SynthID watermark and the enterprise infrastructure it runs on, so readers who need a safety review should look for separate documentation.

Our reading is that this is a real product step, because length, price and general availability are concrete and checkable. The quality claims are not yet checkable. A team that makes product or ad video at volume should price a sample job first: clips needed, seconds per clip, expected retries and the resolution of the final file. It should then run its own reference images and see whether faces, logos and packaging text stay stable, and whether the 15-second clips need the allowlisted audio feature at all. The price cut and the longer clip are useful only if the results survive that test.