Google's Gemini Omni 1.2 Flash Makes 15-Second AI Video Clips You Can Actually Edit

Gemini Omni 1.2 Flash stretches AI video to 15 seconds

Google released Gemini Omni 1.2 Flash on November 5, and the headline number is 15. The new video model generates and edits clips from 3 to 15 seconds, up from the 10-second ceiling on version 1.1. That extra five seconds sounds small until you try to fit a complete shot into a clip. Ten seconds forces you to cut mid-action. Fifteen seconds covers a full sentence, a product turn, or a short scene without a splice.

The model is generally available now through the Gemini API and Vertex AI, and it is rolling out across Flow, Google Vids, and the Gemini app. Google also cut the price. 720p video now costs $0.07 per second, which Google says is 30 percent below version 1.1. A 15-second 720p clip runs about $1.04 before retries.

The real upgrade is conversational editing

The longer clips matter, but the editing workflow matters more. Gemini Omni 1.2 Flash remembers the state of a clip, so you can follow up with an instruction like "change only the background" instead of starting over from a fresh prompt. Each edit produces a newly synthesized video, so this is not a timeline editor. There is no editable timeline, no layers, no keyframes. What you get is a linked, reviewable draft that keeps the same scene while you iterate on details.

That distinction is worth stating plainly. If you need frame-accurate cuts, use a real editor. If you need ten variations of the same shot with different lighting, backgrounds, or outfits, the conversational loop saves real time. Google's demo gallery shows the pattern clearly: one photo of a model plus front and back shots of three outfits becomes a 15-second virtual try-on clip.

The model accepts text, images, audio, and video as inputs, and it can generate synchronized native audio alongside the picture. First-and-last-frame workflows are limited to 3 to 10 seconds, while straight generation and editing cover the full 3 to 15 second range. Output comes in 16:9 and 9:16, from 360p up to 4K with upsampling.

Gemini Omni 1.2 Flash pricing and where you can use it

The price cut is the part most working creators will feel first. Google lists standard output at $12 per million tokens, which works out to about $0.023 per second at 360p, $0.07 at 720p, $0.104 at 1080p with upsampling, and $0.209 at 4K with upsampling. The 720p figure is the one Google announced directly, and it is the tier most people will actually use.

Access depends on where you work. Developers call the model as gemini-omni-1.2-flash through the paid Gemini API tier. Enterprise teams get it through Vertex AI, including a Provisioned Throughput option. On the consumer side, Flow users get the new model now, with 15-second clips, Extend, and upscaling reserved for Google AI Plus, Pro, or Ultra plans. Free Flow accounts stay at 10 seconds. In the Gemini app, 15-second video is rolling out to Google AI Ultra subscribers first.

One catch: audio-reference input remains allowlisted rather than generally open. And editing or extending uploaded video is unavailable in the European Economic Area, Switzerland, and the United Kingdom, although editing model-generated video still works there.

The old models are on a countdown

Google set a firm retirement schedule, so treat this as a migration, not an optional upgrade. Veo Extend was already retired on November 5. Veo 3.1 Fast, Veo 3.1 Quality, and Omni Flash 1.1 are retired in Flow on December 5, 2026. On the API side, requests sent to gemini-omni-1.1-flash will automatically route to 1.2 starting December 3, billed at the lower 1.2 rates.

That automatic reroute sounds convenient, and it is, but it changes output. A prompt tuned for 1.1 will not produce identical video on 1.2. If your workflow depends on a specific look, test your core prompts on 1.2 before December rather than discovering the difference in production.

Google also confirmed that all generated videos carry an imperceptible SynthID watermark, and safety filters apply to prompts and outputs. The company says development continues on fast sports action, rapid script changes, and complex physics, which reads as an honest list of the places the model still struggles.

What this means for the AI video race

The timing is no accident. OpenAI split its image model in two this week and added animated GIF creation to ChatGPT, while Tencent's Hy4 added vision and computer use to its open-weight lineup. Thinking Machines Lab also shipped Inkling 2 Small, a 594B model aimed at long coding jobs. Google's answer is to make video generation cheaper, longer, and iterative rather than chasing a single spectacular demo.

That is a sensible bet. Short AI video stopped being a novelty months ago. The bottleneck now is workflow: how many tries it takes to get a usable clip, and what each try costs. Fifteen seconds plus follow-up edits plus a 30 percent price cut attacks all three at once. The model still cannot make a finished commercial on its own. Continuity, audio, rights, and factual accuracy still need a human in the loop. But for the draft stage, where most of the wasted time and money lives, Gemini Omni 1.2 Flash is now the cheapest serious option from a major lab.

If you already use Flow or the Gemini API, switch your default model now and re-test your prompts. If you are still on 1.1, the clock is running. December 3 is less than a month away.