Gemini Omni 1.2 Flash extends video clips and lowers costs

Google makes Gemini Omni 1.2 Flash available for video generation. The model extends clips to 15 seconds, adds conversational editing, and reduces the price by about 30% on 720p.

Google made Gemini Omni 1.2 Flash generally available on November 5, 2026, with a clear goal: to produce longer videos, easier to edit and cheaper to generate. The model extends clips to 15 seconds, up from 10 previously, and reduces the cost of 720p by about 30%, to nearly 0.07 dollars per second. For teams producing multiple versions, this change matters as much for budget as for production flow.

Des clips plus longs, sans repartir de zéro

Gemini Omni 1.2 Flash supports durations from 3 to 15 seconds, with a default 16:9 format and support for vertical 9:16. Google also offers drafts in 360p or 720p, then upscaling to 1080p or 4K on selected renders. The idea is simple: generate quickly, correct afterwards, and pay for high definition only for what really ends up in production.

The most visible novelty remains conversational editing. A user can modify a clip through multiple exchanges, replace a product or a set, change lighting, adjust weather, or switch format without rewriting the entire scene. The system retains context between requests thanks to a previous interaction ID, avoiding resending the original video with each request.

La cohérence visuelle devient l’argument central

Google emphasizes the stability of characters, clothing, packaging, and brand colors from shot to shot. The model accepts text, images, short videos, and, according to the announcement, up to six reference images per request, with a maximum of three images per character for two characters. This approach mainly targets advertising uses, product demonstrations, and scenes where visual drift ruins the sequence.

In its internal evaluations, Google DeepMind reports that Gemini Omni 1.2 achieves a visual preference rate of 62% versus 38% for Omni 1.1 in text-to-video, then 63% versus 37% on identity consistency with multiple references. These are in-house metrics, but they indicate the direction: less morphing, less drift, and transitions that hold up better. It’s not a miracle, just a video that stops turning into a distant cousin in the middle of the shot.

Ce que les équipes peuvent en faire, et ce qui reste limité

The model is designed for iterative editing. Google describes a loop where a first render is prepared, a specific detail corrected, then only the best version finalized in high resolution. This logic applies especially to product announcements, social clips, artistic direction trials, and scenarios where a full generation costs less than a series of manual retakes.

But limits remain explicit. Rapid script changes, very dynamic sports, and some complex physics effects remain difficult. The API is fully supported in English, with no published evaluation for other languages. Imported voices, reference audio, and voice editing are not supported at this stage [to be verified], and safety filters continue to apply to prompts as well as generated videos.

Une diffusion large, entre API, studio et produits Google

Deployment goes through Gemini API, Google AI Studio, Google Cloud Console, Flow, Google Vids, the Gemini app for Google AI Ultra subscribers, and Google Ads. Google also adds an enterprise layer with Cloud infrastructures, security functions, and Provisioned Throughput capacity. Each video produced via the API receives an invisible SynthID watermark, detectable by program, for provenance verification.

Pricing follows the same industrial logic: 12 dollars per million video output tokens, about 0.70 dollars for 10 seconds and 1.04 dollars for 15 seconds, according to Google’s figures. Text, image, and video inputs are charged 1.50 dollars per million tokens. Gemini Omni 1.1 Flash remains accessible during a four-week transition, then API requests will automatically switch to 1.2 on December 3, 2026, with new rates. For existing customers, the message is clear: no immediate rewrite, but a scheduled migration.

Le pari de Google sur la production vidéo industrielle

This version does not just seek to generate longer clips. It mainly aims to make editing reliable enough to replace some full regenerations. This interests players like Photoroom, whose CEO Matt Rouif says that longer duration was a major request and that general availability with SLA allows them to consider Gemini Omni Flash in production for vendors and enterprise clients. The model already powers their Video Ad Maker and Product Animator tools.

Ultimately, the challenge is this: if a video AI keeps faces, products, and scenes from one version to another, it becomes a production chain tool, not just a demo generator. In this context, Google tries to move generative video from a test ground to scalable campaign use, with known limits but a framework closer to business needs.

Key points

  • General availability announced on November 5, 2026.
  • Clips from 3 to 15 seconds, up from 10 previously.
  • 720p price reduced by about 30%.
  • Google DeepMind cites 62% versus 38% in visual consistency.
  • Photoroom already uses it for Video Ad Maker.
  • Automatic switch from Omni 1.1 on December 3, 2026.

By the numbers

  • 15 seconds — maximum clip duration, worldwide, Google announcement November 5, 2026.
  • 0.07 dollar — approximate cost per second at 720p, worldwide, Google announcement.
  • 12 dollars — price per million video output tokens, worldwide, Gemini API.
  • 62% — visual preference rate versus 38% for Omni 1.1, Google DeepMind internal tests.
  • December 3, 2026 — automatic switch date for API requests to 1.2.

To read