Dek: Two numbers decide whether this model belongs in your production budget: fifteen seconds and seven cents. The rest is workflow, and the workflow is where the money actually moves.
What Gemini Omni 1.2 Flash changes for video teams
Google published the Gemini Omni 1.2 Flash post on November 5, written by Anish Nangia, a product manager at Google DeepMind. The model is now generally available on the Gemini API and Google Cloud, and it also shows up in Gemini for business and in Google Vids.
Strip the launch language and you get three changes that matter to anyone shipping video features. Clips got longer, and iteration got cheaper. Editing stopped requiring a full regeneration.
That's a different pitch from the usual "better model" story. Google isn't selling you a new quality ceiling here. It's selling a lower cost of being wrong, which is the thing that actually limits how much video a small team can produce.
15-second clips, set in one-second steps
Native clip length moved from 10 seconds in Gemini Omni 1.1 to 15 seconds. Duration is now settable in one-second increments, anywhere between 3 and 15 seconds.
The one-second step sounds like a spec-sheet detail. It isn't. Ad slots come in fixed lengths, and a 6-second bumper or a 12-second cutdown usually has to be trimmed by hand when the generator only offers round numbers. Being able to ask for the length you need removes a step that used to land on an editor's desk.
Fifteen seconds also covers a full product beat. You can open on the object, move it, and land a line of copy without stitching two generations together and hoping the lighting matches.
Is that enough for a story? No. It's enough for a shot.
$0.07 per second and the math of a campaign
Google cut the price by 30% against Gemini Omni 1.1, to $0.07 per second for 720p video.
Run the arithmetic on a modest campaign. Ten concepts at 15 seconds each is 150 seconds of generation. At seven cents a second, that's $10.50 for a first pass.
Ten dollars and change.
Before the cut, the same exploration cost noticeably more, and teams rationed it. Now the constraint is taste, not budget. The catch is resolution. Google's stated price covers 720p, and the post gives no figure for 1080p or 4K, so don't build a cost model that assumes the cheap rate holds all the way up the ladder.
Multi-turn editing changes how a take gets finished
Multi-turn editing is the change I'd watch most closely. Instead of rewriting the prompt and regenerating the whole clip, you send a follow-up instruction and the model fixes that detail.
Anyone who has generated video knows the tax here. You get a take that's 90% right, one sleeve is the wrong color, and the only tool you had was the reroll button. Ten generations later you've lost the take you liked and the seed that produced it. Conversational edits turn that into a two-step job: keep the take, change the sleeve.
Reference adherence got attention too. You can combine character, product or style images with short motion clips of up to three seconds, plus audio cues as reference input. Google says the model holds faces, garments, packaging typography and brand colors steady across camera angles. For a retailer with a fixed brand palette, that's the difference between a usable asset and a reshoot.
Note the wording on audio. These are reference cues, not generated sound, so nobody is claiming the model scores your spot.
Gemini API video: draft cheap, upscale the survivors
The workflow Google recommends is the one that makes the pricing make sense. Draft at 360p or 720p, edit with follow-ups, then upscale only the takes that survive review to 1080p or 4K.
That ordering is the product. Cheap drafts plus cheap edits means the expensive step runs on maybe two clips instead of twelve. Teams already work this way with stills, so the habit transfers.
Google names the target jobs directly: virtual try-on clips, catalog spins, branded video ads, product explainers and social ad spots. Look at that list and the buyer is obvious. It's a commerce or performance-marketing team that needs forty variants of the same product shot for forty placements, not a studio chasing a film look.
If your team bills through Google Cloud, the spend lands on the same invoice you already reconcile. For agencies, that's a client-billing question as much as a technical one.
What the Gemini Omni 1.2 Flash post doesn't say
Three gaps are worth flagging before you commit.
There are no quality benchmarks in the announcement. No side-by-side scoring, no preference study, nothing that lets you compare this model against Gemini Omni 1.1 on output quality, which leaves duration, price and editing control as the only claims you can check before you spend anything.
There's no free-tier claim either. Don't assume the API has a trial allowance because earlier models did.
And there's no price for high-resolution output. If your delivery spec is 4K, the seven-cent figure tells you about the middle of the pipeline, not the end.
Where Gemini Omni 1.2 Flash lands for buyers
Cheap iteration is the whole product. A model that lets you throw away nine takes and keep the tenth changes how a small brand plans a quarter, because the cost of exploring stops being the thing that limits it.
Gemini Omni 1.2 Flash is available now through the Gemini API and Google Cloud. Start with one product you already shoot regularly, generate the draft at 720p, and see whether the edit loop holds up on a real brief before you move anything else over.
