A short product teaser sometimes needs one more beat: show the object, let something happen, give the ending room to land. Google’s Omni 1.2 Flash, announced on 5 November, raises native video generation from ten to fifteen seconds. The practical offer depends on which Google product you open.

Based on 7 reviewed sources · See sources

Three clips, or a metered workflow

Google’s Vids announcement offers anyone with a Google account or Workspace subscription up to three 15-second videos per month at no cost. That gives an occasional creator a small allowance for trying a scene or teaser.

Developers get a different deal. The Gemini API is paid: Google’s pricing page lists an effective 720p video-output rate of about $0.07 per second, which Google says is 30% below 1.1. Its more precise token calculation puts a fifteen-second output at about $1.04, before input and text-output charges. A lower rate per second does not mean a longer clip has a lower total bill.

The launch post also distinguishes the Gemini app: Omni 1.2 rolls out to Google AI subscribers, while fifteen-second creation and editing are specified for Ultra. A free Vids allowance, a Gemini subscription and API billing are separate routes.

The detail

What fifteen seconds costs at different resolutions

Gemini API Standard video output; Google’s token-based rates, checked 6 November 2026.

What fifteen seconds costs at different resolutions. Gemini API Standard video output; Google’s token-based rates, checked 6 November 2026.
Output routeApproximate output charge
360p draft · 15 seconds$0.35
720p · 15 seconds$1.04
1080p with upsampling · 15 seconds$1.56
4K with upsampling · 15 seconds$3.13
Calculated from listed video tokens per second × 15 × $12 per million tokens, rounded to cents. Input, text output and additional generations cost extra. Upsampling enlarges an output; it is a separate cost route. These are not Vids subscription prices.Source: Gemini API pricing · 2026-11-06

A longer scene is a different test

Google demonstrates conversational revisions, such as changing a background or subject while continuing work on a clip. Its original announcement below is a vendor demonstration; the claims about steadier detail and more natural speech still need testing against the work you actually make.

From the source · Google ·

Google announces longer video generation, conversational editing and lower API pricing; the attached footage is its demonstration.

View the original post on X ↗

DesignArena, a visual-model comparison platform made by Intelligence, published its own paired Colosseum example. It describes the same requested camera journey for both versions and supplies a comparison video. The reported render times were 88 seconds for 1.2 and 53 seconds for 1.1, but the outputs were fifteen and ten seconds long. That is useful evidence of one task, not a matched-duration speed benchmark.

DesignArena judged the new version’s architecture more consistent. The post does not establish repeatability across multiple attempts. Its roughly equal total charges—$1.04 and $1.01—also illustrate the billing distinction: more footage can absorb the saving per second. For a production decision, the next useful comparison keeps duration, resolution and requested action fixed, then counts rejected attempts.

Your own footage is the important European boundary

The API guide says users in the EEA, Switzerland and UK cannot edit or extend uploaded videos. Editing or extending videos generated by the model is supported. For a Belgian team, generating a new scene and asking the tool to alter an existing shoot are therefore materially different jobs.

The Vids help page similarly excludes uploaded-video editing in those regions, allowing edits to AI-generated clips in generation history. At the time checked, that page still described three-to-ten-second controls, despite the new fifteen-second announcement. Treat the announcement as the new offer and check the controls in your account before promising a particular workflow.

Some API paths also keep a shorter limit: supplying a first and last frame supports three-to-ten-second clips. The guide does not currently support audio-file inputs or audio references. A model demonstration is not a guarantee that every input route is available in every product.

Sources

Written with AI assistance from the sources above. Analysis and practical implications are our interpretation. Our editorial approach.