Gemini Omni 1.2 Flash Extends Video Clips to 15 Seconds—and Puts Revision at the Center

Chinese

Google has made Gemini Omni 1.2 Flash generally available, with native 15-second clips and lower 720p pricing. The practical question is whether conversational revisions reduce the cost of an approved video, rather than merely the price of a generation.

AI-generated concept illustration: paper landscapes and a recurring silver sphere suggest continuity and revision.

For a small video team, the expensive moment often comes after the first draft: the scene works, but the product, background or lighting needs changing. Google’s November 5 release of Gemini Omni 1.2 Flash targets that revision loop, alongside longer clips and a lower generation price. This is an update to an existing model family, not its first appearance. [1]

Google says the model can generate and edit native clips lasting up to 15 seconds, compared with 10 seconds in Omni 1.1. It supports follow-up instructions on the same clip. The company also describes improved adherence to references and more consistent characters and objects. Those quality claims and the published demonstrations come from Google; AIYXL has not reproduced them. [1]

The release is a service milestone as well as a feature update. Google Cloud’s November 5 announcement identifies the model as generally available and offers Provisioned Throughput. Its quoted 720p price is $0.07 per generated second, 30% below Omni 1.1 Flash. A 15-second output therefore has a $1.05 generation price at that rate before retries, further edits, upscaling or other applicable charges. That calculation is a budgeting illustration, not a measured production cost. [2]

Availability needs to be read at the product level. Google announced access through developer and business channels, while the Gemini app rollout starts with Google AI Ultra subscribers. Android Authority independently reported the announcement and the differing access routes on November 5; its coverage is corroboration of the release, not an independent quality benchmark. General availability does not mean every account has the same features or quota. [1] [4]

There are narrower controls inside the broader launch. Google Cloud says audio references require allowlisting. The DeepMind model card also says a capability to alter people’s speech is being restricted while the company studies how to make it available responsibly. These qualifications matter when planning a workflow around a demonstration: visible model capability and permission to use it are separate questions. [2] [3]

Editorial analysis: the useful comparison for a studio is cost per approved deliverable. A lower per-second price helps, but a revision that silently changes a package label or a face can erase the saving. A sensible pilot would keep the brief and references fixed, record each attempted change, and count how often previously approved details survive. Nothing in the launch materials establishes that result for an individual customer.

Fifteen seconds gives a creator more room for a product explanation or a short scene. The stronger test comes on the next instruction: whether changing one detail leaves the rest intact. That is where this release could reshape everyday production—and where its claims still need to meet a team’s own acceptance checks.

Sources:

  1. Google product announcement — 2026-11-05.

  2. Google Cloud launch announcement — 2026-11-05.

  3. Google DeepMind model card — 2026-11-05.

  4. Android Authority reporting — 2026-11-05.