Fifteen seconds changes what the model is for
Google made Gemini Omni 1.2 Flash generally available on November 5, 2026, and the headline number is a small one: clips can now run up to 15 seconds, up from 10 on Omni 1.1. Five extra seconds sounds like a footnote. It isn’t. At 10 seconds, an AI video model is a shot generator — a single moment, a single camera move, a single idea. At 15 seconds, with the new multi-turn editing, it becomes something closer to a scene: enough room for a setup and a payoff, a product demo with a beginning and an end, a dialogue exchange instead of a single line.
The length is selectable in one-second increments from 3 to 15 seconds, so you’re not locked into the maximum. That granularity matters more than it sounds like, because the pricing is per second, and a 4-second clip costs a fraction of a 15-second one. Google is selling precision here as much as length.
This is a genuinely available product, not a waitlist announcement. It’s live across the Gemini API and AI Studio, Vertex AI, Flow, and Google Vids, with the Gemini app rollout going to Google AI Ultra subscribers and availability in Google Ads. The model ID is gemini-omni-1.2-flash, and from December 3, requests to the old gemini-omni-1.1-flash will automatically route to 1.2 and bill at 1.2 rates. If you’re still calling 1.1, the upgrade is coming whether you ask for it or not.
The price, worked out per clip
The other headline number is the price: $0.07 per second at 720p, which Google says is 30% less than Omni 1.1. The underlying token price tells the same story with more precision — $12.00 per million video-output tokens versus $17.50 for 1.1, a 31.4% cut that Google rounds to 30%. At 720p the model uses 5,792 tokens per second, and text input runs $1.50 per million tokens.
What does a clip actually cost? At 720p, roughly $0.35 for 5 seconds, $0.70 for 10 seconds, and about $1.04 for a full 15 seconds. That’s the video-output cost; reference images and text input add a little on top, but the per-second figure is the number to remember. Other resolutions price differently, per Google’s pricing page: around $0.023 per second at 360p, about $0.104 at 1080p, and roughly $0.209 at 4K, which puts a 15-second 4K clip at about $3.13.
There’s no free tier on the Gemini API, so every clip costs something — but at these rates, the cost of experimenting is trivially small. The meaningful number isn’t the per-clip price, it’s how many clips you burn iterating toward one you keep. Which is exactly why the next feature matters more than the price cut.
Draft cheap, upscale the keepers
Omni 1.2 Flash formalizes a workflow that experienced video-AI users were already doing by hand: draft at low resolution, upscale only the winners. Generate at 360p or 720p, then upscale the best clips to 1080p or 4K. The upscaling step is built into the model now, and the price ladder above makes the economics obvious — a 720p draft costs a third of the 4K render, so you can afford to be wrong several times before committing to the expensive version.
This is the part of the release that separates a tool you play with from a tool you build a workflow around. Drafting at 360p for roughly two cents a second means a creator can generate twenty variations of a shot for less than the cost of one finished 4K clip. The bottleneck stops being money and becomes taste — which variations are worth upscaling — and that’s a much better problem to have.
Talk to the same clip until it’s right
The other half of the release is multi-turn conversational editing. You can now keep working on the same clip across turns — swap the product, change the background, adjust the lighting or weather, change the aspect ratio — and the model holds the scene together across the full 15 seconds. Google also claims stronger reference following: feed it character, product, or style images, motion clips up to 3 seconds, or audio cues, and it keeps them consistent through the clip.
Multi-speaker dialogue and lip-sync are also improved, according to Google. These are the maker’s claims, and there are no independent benchmarks published yet, so the honest way to read them is as a direction of travel rather than a measured result. The claims are specific enough to test, though: same character across a 15-second scene, two speakers talking without the lip-sync falling apart, a product that stays the same product from the first frame to the last. Those are the things to check in your first week with the model, because they’re also the things that were hardest in the previous version.
Audio references are gated behind allowlisting, per Google’s own documentation, so not every account will have that on day one. First-and-last-frame workflows are reported to be limited to 3–10 second clips, and up to 6 reference images are supported — both from secondary sources rather than Google’s own pages, so treat them as reported rather than confirmed.
Where you can actually use it
Access depends on where you work. Developers get it through the Gemini API and AI Studio and through Vertex AI. Flow, Google’s AI filmmaking tool, has it built in. Google Vids users get 15-second clips at no additional cost with a Google account or Workspace account, which is the easiest on-ramp if you just want to try it without touching an API key. The Gemini app is rolling it out to Google AI Ultra subscribers, and it’s available in Google Ads for advertisers generating creative.
Two things to know before you start. First, outputs carry an imperceptible SynthID watermark, so generated video is marked as AI-made even when it looks real. Second, Provisioned Throughput is available for teams that need guaranteed capacity rather than pay-as-you-go. And if you’re migrating from 1.1, remember the December 3 auto-routing date — your old model ID will silently start billing at the new rate, which is lower, but it’s still a change worth knowing about.
The fine print worth knowing
Omni 1.2 Flash is a strong release, but the honest version includes what Google hasn’t shown. There are no independent quality benchmarks, so the consistency and lip-sync improvements are Google’s word. The model is a Flash model — fast and cheap — which means it’s not Google’s highest-quality video model, and there will be shots it can’t hold together that a slower, more expensive model could. And the retirement date floating around in secondary coverage — November 5, 2027 or later — comes from a single secondary source citing the docs, so don’t plan around it.
Who is this for? Creators making short-form video who want a full scene rather than a single shot; advertisers and marketers generating product clips at a price where iteration is nearly free; developers building video features into apps where $1.04 for a 15-second 720p clip makes the unit economics work. Who should skip it? Anyone who needs long-form video — 15 seconds is still 15 seconds — and anyone who needs guaranteed character consistency for a brand campaign, until the multi-turn editing has been tested in the wild for a few weeks.
The price cut is real, the 15 seconds are real, and the draft-then-upscale workflow is the part that will actually change how people use it. The rest is a claim waiting for its first real test.
Also today: ChatGPT Images 2.6 adds looping GIFs — and quietly raises the cost of cheap images and Agent-browser: the CLI that gives AI agents a browser. For more on the AI model race, see our GPT-6.1 Terra and Meta Muse Spark Open coverage.