Google on 5 November 2026 made Gemini Omni 1.2 Flash generally available on the Gemini API and Google Cloud. It can now generate and edit clips up to 15 seconds, up from 10 seconds in Omni 1.1. Google set 720p generation at $0.07 a second, which it says is 30% less than Omni 1.1.
That is the change for teams already rendering product clips or short ads in Google’s stack. Google published no independent quality scores, and several modes still stop short of 15 seconds.
Longer clips, with mode-by-mode limits
Cloud documentation for the model id gemini-omni-1.2-flash matches the product blog’s 3 to 15 second range for generation and editing, and lists output at 360p, 720p, 1080p, and 4K. Aspect ratios are 16:9 and 9:16.
Other workflows are stricter. First-and-last-frame generation runs 3 to 10 seconds, and video extension adds 3 to 10 seconds onto up to the last 30 seconds of the input. A prompt can include up to 10 images and 3 videos. Motion reference clips in the product blog are “up to three seconds.”
Google’s Cloud blog says audio references are gated behind allowlisting. The docs say the model can output synchronized speech, music, and sound effects. The DeepMind model card notes that it can change people’s speech when editing video, and that Google is restricting that capability for now.
Price, drafts, and what Google claims improved
The $0.07 a second figure is for 720p. Google says there are similar savings at other resolutions, without listing those rates. The suggested workflow is to draft at 360p or 720p, fix a detail with a follow-up edit instead of regenerating the whole clip, then upscale keepers to 1080p or 4K.
Google says Omni 1.2 Flash holds character, product, and style references more tightly across camera moves than 1.1, with less visual drift. It also claims more natural dialogue, clearer lip-sync, and steadier camera moves. Those comparisons come from Google. We have not seen an independent side-by-side, and the company says work remains on fast sports action, rapid script changes, and complex physics.
Multi-turn conversational editing is the change underneath that quality copy. Google says you can swap a product or background, change the lighting, or switch aspect ratio on the same clip without starting over. Days after OpenAI added GIFs, SVG export, and faster image edits for ChatGPT stills, Google is pushing short, reference-driven video clips at catalog scale.
Where it shipped, and how outputs are marked
The model is generally available on the Gemini API and Google Cloud, with Provisioned Throughput on Cloud. Outputs carry an imperceptible SynthID watermark. Cloud docs also list Content Credentials (C2PA) as supported. That matters for teams selling into the EU, where older generative tools must add machine-readable marks by 2 December.
Google Vids began a full rollout on 5 November 2026 across Business, Enterprise, Education Plus, and consumer Google accounts. Clip lengths there run from 3 to 15 seconds, and the Workspace post says the feature has no admin control.
Google’s product blog says 15-second clips in Vids come at no additional cost with a Google account or Workspace subscription. In the Gemini app, 15-second video is rolling out to Google AI Ultra subscribers.
What to do this week
If you already generate short product or ad clips through the Gemini API or Vertex, point a test job at gemini-omni-1.2-flash and compare a 10-second clip against a 15-second one at 720p, using the same references. Check whether your workflow hits a 10-second cap (first and last frames, or extension) before you promise 15-second output to a client. If you publish in the EU, confirm that SynthID and C2PA metadata survive your edit stack before the 2 December marking deadline.
Sources
Get the weekly email
The week's most useful AI news, a tutorial and an interview.
Lars Strother
Lars Strother is the senior editor for models and hardware at AI Observer, covering models and research, chips and hardware, and robotics. He came to journalism from engineering and still reads the benchmark tables before the press release. His coverage is measured and skeptical, and he wants to know what was tested, on what, and against which baseline.
Spotted a mistake? Tell us on the corrections page.