Google's AI video push took its biggest step yet on Thursday: Gemini Omni 1.2 Flash is now generally available across the Gemini API and Gemini for business. For the first time, the Omni line is a production product rather than a preview — backed by Google Cloud infrastructure, with reserved-capacity Provisioned Throughput for teams that need it, and an imperceptible SynthID watermark on every output.

The headline changes are length and price. Clips now run natively to 15 seconds, up from 10, with duration selectable second by second from 3 to 15. And the 720p rate falls 30% to seven cents a second — which puts a full-length 15-second draft at about a dollar. Google's suggested workflow is to draft cheap at 360p or 720p, fix details with follow-up edits instead of regenerating, and upscale only the keepers to 1080p or 4K.

Reference following is the real upgrade#

The more consequential change is how well the model holds onto what you hand it. You can feed it character, product or style images, short motion clips of up to three seconds, and audio cues, and it keeps faces, garments, packaging type and brand colours consistent across angles and cuts. Google's demo case makes the point concrete: one model photo plus front-and-back shots of three outfits becomes a 15-second virtual try-on sequence, with 360-degree turns and match cuts, while the model and each garment hold steady.

That matters because consistency — not raw clip length — is what has kept AI video out of real advertising work. A model that forgets the product halfway through a shot is a demo, not a tool. Omni 1.2 Flash also promises stronger frame-to-frame consistency in general: less visual drift or morphing across motion and camera transitions, and steadier results when interpolating between a first and last keyframe or extending a clip into a longer sequence.

The same young man holding the same bottle, shown identically across a row of film frames on a desk
AI-generated editorial illustration.

Editing in conversation, not from scratch#

The second shift is workflow. Instead of starting over when a detail is wrong, you refine the same clip turn by turn — swap a product or background, change the lighting or weather, switch the aspect ratio. Multi-turn editing works across full 15-second clips, so one generated scene can spin out ad variants for different audiences and placements without re-briefing the model each time.

Speech gets more natural too: more fluid dialogue, clearer lip-sync, and distinct voices per speaker in multi-character scenes — aimed at dialogue in Flow, explainers in Gemini for business, and narrated clips in Google Vids. Camera work is steadier as well: continuous moves such as sweeping FPV flythroughs, dramatic zoom-outs and orbits, with fewer unwanted jump cuts. Google's showpiece is a single continuous FPV drone flight through the Pantheon, from the piazza through the bronze doors and up through the dome to the oculus, with no cuts at all.

A camera drone flying inside a grand domed hall with columns, light streaming through the oculus
AI-generated editorial illustration.

Where to use it — and what it still can't do#

Developers can prototype in Google AI Studio; businesses get it in the Cloud console and API; and inside Google's own apps it lands in Flow, Google Vids, the Gemini app, and Google Ads campaign tooling. The most accessible on-ramp is Vids: anyone with a Google account or Workspace subscription can generate up to three 15-second videos a month at no cost. In the Gemini app, 15-second generation is rolling out to AI Ultra subscribers first.

Google is candid about the limits: fast sports action, rapid script changes and complex physics remain weak, and the company says work there is ongoing. And the usual caveat applies to everything above: the consistency and quality claims come from Google's own demos and write-up — no independent lab has published hands-on results yet.

What to watch#

Fifteen seconds is a threshold, not just a number: a complete product explainer, a full social ad spot, or a short with a beginning, middle and end now fits in a single generation. The question is whether that, plus conversational editing, is enough to move real ad production onto the model — and whether rivals answer on price. Watch the Gemini app rollout too: until 15-second generation reaches beyond Ultra subscribers, most consumers will meet the model through Vids, not the app.

Sources: Google's announcement on the Google blog; FoneArena; Dev.to; and TSN Media. Images are AI-generated editorial illustrations.