Videos default to 720p in a widescreen 16:9 format, although vertical 9:16 output is also supported. Google is positioning the model for businesses and creators producing advertisements, product demonstrations and other short-form video content.
One of the most significant additions is the ability to revise existing videos through a series of conversational instructions. Rather than generating an entirely new clip whenever something needs to change, users can request specific adjustments while preserving other elements of the original scene.
For example, a creator can replace a featured product, modify the background, adjust weather conditions or change the lighting without rebuilding the entire sequence. The model can also convert footage between landscape and portrait formats, providing additional flexibility when preparing content for different platforms.
Gemini Omni 1.2 Flash also introduces stronger support for reference materials. Users can combine images of characters, products and visual styles with short motion clips and audio cues to guide the generation process. Motion references can be up to three seconds long, allowing creators to provide examples of movement alongside visual instructions.
Google says the updated model better preserves details such as facial appearance, clothing, product packaging and brand colors across changes in camera perspective. These improvements are intended to reduce inconsistencies that can make AI-generated footage difficult to use in commercial production.
In internal evaluations conducted by Google DeepMind, Omni 1.2 Flash outperformed its predecessor in two measures of visual consistency. Evaluators preferred the newer model 62% of the time for text-to-video consistency and 63% of the time when assessing whether identities remained consistent across multiple references.
The company has also added a video extension capability that allows users to build on previously generated footage. Each extension adds between three and 10 seconds, with a maximum combined duration of 30 seconds. The system uses the preceding clip's final 10 seconds as context to continue the sequence.
The pricing changes are another major part of the release. At $0.07 per second, a 10-second video generated at 720p costs approximately $0.70 through the paid Gemini API, while a 15-second clip costs about $1.05. Google says comparable price reductions apply to other supported resolutions.
The model is not available through the Gemini API's free tier. Developers currently using Gemini Omni 1.1 Flash will also be moved to the newer version automatically on December 3, 2026, with existing API requests redirected without requiring code modifications.
Gemini Omni 1.2 Flash is accessible through Google AI Studio, the Gemini API and Google Cloud's enterprise services, including Provisioned Throughput. The model is also being introduced across Google's consumer and business products, including Flow, Google Ads and Google Vids.
In Google Vids, users with a Google account or Workspace subscription can generate 15-second clips without an additional charge. The Gemini app is separately rolling out the capability to Google AI Ultra subscribers. Access to audio-reference features through Google Cloud currently requires allowlisting.
Google has identified several companies already working with the updated model, including Picsart, Photoroom, invideo, Krea, Artlist, FLORA, Runway and QuickFrame AI.
Photoroom is using Gemini Omni Flash for its Video Ad Maker and Product Animator, two tools designed to help businesses create promotional content. The company's CEO and co-founder, Matt Rouif, cited the model's combination of output quality, generation speed and pricing as reasons for selecting it.
"We chose Gemini Omni Flash to power our Video Ad Maker and Product Animator because it balances quality, speed and cost," Rouif said.
QuickFrame AI has highlighted the ability to produce 15-second footage at 1080p, allowing users to generate a connected-TV advertisement within a single video creation process. Picsart, meanwhile, plans to extend access to the model across its creator platform.
"Now that it is generally available, we can bring this quality to over 100M Picsart creators," said Narek Hayrapetyan, Picsart's head of product and partnerships.
The commercial focus reflects Google's effort to make AI video generation more practical for companies producing content repeatedly. Short advertisements, virtual product demonstrations and promotional videos often require multiple versions, making generation costs and editing flexibility important considerations.
Despite the improvements, Google acknowledges that certain video generation tasks remain challenging. The company is continuing development work on rapid athletic movements, abrupt changes in scripted action and scenes involving complicated physical interactions.
There are also limitations in the current API. English is the only language Google has fully evaluated, while voice editing and user-uploaded audio inputs are not yet supported. Audio references are available only through approved Google Cloud access.
To help identify AI-generated content, Google embeds an invisible SynthID watermark in every video produced by the model. The watermark can be detected using software designed to verify content provenance.
Gemini Omni 1.2 Flash builds on Google's earlier Omni Flash release, introduced at its I/O conference in May, which initially supported videos lasting up to 10 seconds. The latest version combines longer generations with more granular editing, improved reference handling and reduced API pricing.
For developers and businesses already using Google's AI video tools, the release offers a way to produce longer clips at a lower cost while making targeted revisions without repeatedly starting over. Its broader availability across Google Cloud, creative applications and the Gemini ecosystem also gives the company additional channels for bringing AI-generated video into existing production workflows.
This analysis is based on reporting from TNW.
Image courtesy of Google.
This article was generated with AI assistance and reviewed for accuracy and quality.
