Google says Gemini Omni 1.2 Flash is available at $0.07 per second for 720p, and calls that a 30% cut from Omni 1.1 Flash. Multiply that rate by the new 15-second native ceiling and you get $1.05. That figure is arithmetic. Google did not print it as a package price in the launch post, Introducing Gemini Omni 1.2 Flash. Fifteen seconds is a length media buyers already buy. The rate is something a finance partner can put in a spreadsheet. The rest of the card is not filled in yet.
Length and price are the easy headlines. The sharper question is which seconds you are actually paying for, and what leaves the building when you generate them.
Fifteen seconds, one second at a time
Gemini Omni 1.2 Flash is the newest model in Google’s Omni video line. In that same post, Google says it is generally available on the Gemini API and through Gemini for business. Native creation and editing both cap at 15 seconds, up from 10 seconds in Omni 1.1. Duration is not a fixed bucket. You set it in one-second steps from 3 to 15 seconds.
That dial is easy to skip. Don’t. A six-second bumper and a fifteen-second spot are different products. Ask for the runtime you actually need, instead of trimming a longer generation or stretching a short one. That dull control is what shows up on a production calendar.
What you can attach, according to Google: character, product, or style images, motion clips of up to three seconds, and audio cues. What Google claims those references preserve, in its own demos rather than a scorecard: faces, garments, packaging type, and brand colors hold from the first frame to the last, with less drift and less morphing. Dialogue is supposed to sound more fluid. Lip-sync should be clearer, and voices should not bleed into each other. Camera moves are supposed to stay continuous, with fewer unwanted jump cuts and steadier scale.
Hold those as product claims. The launch post does not include benchmark scores for this version. Demos are a pitch. A pilot on your own packaging is a test.
A lower rate is not a cheaper workflow
A 30% cut on 720p is simple to quote. Google’s sharper argument is a sequence, and it is not a finished discount. Draft at 360p or 720p. Fix one wrong detail with a follow-up edit instead of regenerating the clip. Then upscale to 1080p or 4K only after you have a take worth keeping. Google says multi-turn editing works across the full 15 seconds, including swaps of product, background, lighting, weather, and aspect ratio.
Too many teams still use video models like a slot machine. Generate, wince, generate again, and pay for the wince. If a conversational fix is reliable, the savings can stack past the rate card. If it is not, you are back to full rerolls, just at a lower per-second price for 720p.
Do not book that path as a known saving. Google states the 30% reduction and the new 720p price. The launch post does not print the old rate, and it does not publish prices for 360p, 1080p, 4K, edits, or upscales. A clean 30% cut to seven cents implies ten cents before. That is inference, not a quoted predecessor price. An edit is not automatically cheaper than a reroll. 360p is not automatically the bargain tier, either, until those rates exist.
Treat the workflow as a test design. On one real brief, log edit calls against full regenerations. Count dollars only where Google has published a rate. Do not rewrite the creative budget on resolutions the announcement never priced.
Where a name is not a switch
Availability is one map, taken from Google’s post, and then it stops.
The Gemini API and Gemini for business are described as generally available. That path includes Google AI Studio, the Google Cloud console, and an enterprise API framed around enterprise-grade infrastructure and security. Provisioned Throughput is available for teams that need reserved capacity. The post does not spell out the commercial terms. The name is a capacity option, not a finished contract.
Google Vids offers 15-second clips at no additional cost if you have a Google account or a Google Workspace subscription. That is included access. Not a free public faucet.
In the Gemini app, access is rolling out to Google AI Ultra subscribers. Rolling out is not the same sentence as “everyone has it this week.”
Flow and Google Ads are named. A name-check is an announcement. It is not proof a buyer can brief the model there today.
Outputs carry an imperceptible SynthID watermark, Google’s marker for synthetic media. That is provenance. It is not a substitute for disclosure rules, and it will not satisfy every legal or brand policy on its own.
What leaves the building
The path Google describes is a hosted API and Cloud service. You do not get a download you can stand up from the announcement. For a pilot, that is a data question before it is a picture-quality question.
Product shots, talent faces, unreleased packaging, and scripts leave your environment when you send them. Do not upload anything you would not put in a vendor’s cloud on a standard review: an unreleased SKU, identifiable talent without a release that covers synthetic use, or a script still under legal hold. Before a client-facing test, ask where prompts and reference files are stored, who can see them, and how long they persist. Picture quality can be judged on dummy assets. Data handling cannot.
Use it where the shot is supposed to be boring
The honest fit is the shot you already know how to make and are tired of remaking, not the highlight reel. A product on a table. A background swap for a regional campaign. The same configuration, once for a slide and once for a phone, after legal changes a single word in a locked script.
Sales feels this in a specific shot, not a slogan. A CRM copilot does not hand a rep a six-second, on-brand clip for a follow-up. This model might, if the references hold and packaging text stays packaging text while the setting changes. Call it a production bet. Do not mistake it for a media-buying system.
Google also says where the model still falls over. Fast sports action, rapid script changes, and complex physics remain under development. DeepMind’s broader Omni page markets an intuitive grasp of gravity, kinetic energy, and fluid dynamics. Prefer the version note over the family slogan. A football spot, a tricky pour, or a script that keeps rewriting itself mid-sentence is a bad first bet.
There is a production lesson in that limitation. “Rapid script changes” is a process failure as much as a model failure. Approve the line, then generate. Spend the iterations on framing, product, and setting (the edits Google is explicitly selling) instead of treating the model as a live rewrite engine.
Questions worth asking before you brief it
- How long can a clip be, and what does 720p cost?
Native generation and editing cap at 15 seconds, in one-second steps from 3 seconds. Google lists $0.07 per second for 720p and calls that 30% below Omni 1.1 Flash. Fifteen seconds at that rate is $1.05 of arithmetic, not a listed bundle, and the post does not print the prior rate. - Is the quality jump measured?
Not in the launch post. Claims about consistency, speech, lip-sync, and camera stability come from Google and its demos. There are no version 1.2 benchmark scores there to separate a real gain from a better reel. - Where can a team use it today?
Google says the Gemini API and Gemini for business are generally available, including AI Studio, Cloud, and an enterprise API. Vids includes 15-second clips with a Google account or a Workspace plan. Gemini app access is still rolling out to AI Ultra. Flow and Google Ads are named, not confirmed as places you can brief it today. - How do you keep the bill from drifting?
Log edit calls against full rerolls on one brief. Google recommends a draft, then a detail fix, then an upscale of the keeper. It has not priced 360p, edits, upscales, 1080p, or 4K in the launch post, so do not assume those steps are cheaper than a 720p reroll. - What should you not trust it with yet?
Google says fast sports action, rapid script changes, and complex physics still need work. Brand-stable product shots and locked scripts are a better first test than physics-heavy spots. Keep unreleased packaging, talent, and held scripts off the pilot.
The teams that get value here will not wait for a leaderboard this release does not provide. They will pick one repetitive asset, run the draft-edit-upscale path against a full regenerate, and count how often the logo, the face, and the spoken line survive. If those three hold, $0.07 a second is a production tool. If they do not, it is still a demo, just a cheaper one at 720p.