Gemini Omni 1.2 Flash launches: 15-second video and multi-turn editing

Google DeepMind announced the Gemini Omni 1.2 Flash video generation model on November 5, 2026. The model is generally available through the Gemini API and Google Cloud, with support for generating longer clips, better adherence to reference images, and editing the same video through conversational turns. Google lists the price of 720p video generation at $0.07 per second, 30% less than Gemini Omni 1.1.

How are clip length and generation costs changing?

The maximum video length has increased from 10 seconds in the previous version to 15 seconds in Gemini Omni 1.2 Flash. Generation length can be set from 3 to 15 seconds in one-second increments. According to Google, this makes it possible to create longer narratives, product showcases, or social media ads in a single generation.

Initial drafts can be generated at 360p or 720p. Outputs users like can then be upscaled to 1080p or 4K. The company says costs can be reduced not only through lower prices, but also by submitting a follow-up editing prompt for an existing video instead of regenerating the entire clip to change a detail.

Reference inputs and editing

The model can use character, product, or style images alongside short motion clips of up to three seconds and audio cues. Google says it aims to preserve these references more consistently throughout a video and reduce visual changes during scene or camera movements. The company’s examples highlight how the same character, product, and scene can be maintained from different angles.

After generating a clip, users can make changes through conversational turns, such as changing the background, adjusting the lighting or weather, or changing the aspect ratio. This eliminates the need to prepare a new initial prompt for every variation. Google also mentions smoother speech, lip synchronization, and the ability to direct multiple speakers.

Examples of camera movements provided by the model include continuous fly-throughs and push-ins. In a scenario illustrated by Google using architectural reference images, the camera makes a single fly-through from a plaza toward a doorway and into a dome. These are examples shared by the company; its announcement does not provide a numerical success metric showing how results vary with different prompts and references.

Impact on architecture and visualization workflows

For architecture firms and visualization artists, the 15-second generation limit could be useful for creating short spatial narratives or presentation previews. Using reference images together with motion descriptions may help when testing ways to preserve a design language or a particular scene composition. Conversational editing could also speed up the creation of alternatives by changing elements such as the background and lighting.

However, these features do not prove that the model will perfectly preserve the desired architectural geometry or material appearance in every project. Consistency with references and motion continuity should be checked before production. Teams using the API should assess costs based on resolution and clip length; the listed price is $0.07 per second for 720p. Google says 15-second clips can be generated in Google Vids at no extra cost.