Gemini Omni 1.2 Flash creates 15-second videos in one go and cuts 720p costs by 30%
Google has generally released Gemini Omni 1.2 Flash, a video generation and editing model. It supports clips up to 15 seconds and conversational editing, while reducing 720p video pricing by 30% versu
Google officially launched its video generation and editing model, Gemini Omni 1.2 Flash, on November 5, 2026. The most noticeable change is that the maximum length of videos that can be created or edited in one go has increased from 10 seconds to 15 seconds. Users can specify a length from 3 to 15 seconds in one-second increments, while 720p output now costs $0.07 per second, 30% less than 1.1 Flash. This reduces the need to stitch together multiple scenes for short advertisements or product explainer videos.
Make follow-up edits without starting over
The new model supports conversational edits to the same clip. Elements such as the product, background, lighting, weather, and aspect ratio can be changed through follow-up instructions, unlike workflows that regenerate the video from scratch for every minor revision. The proposed production workflow is to quickly create a draft in 360p or 720p, adjust the necessary parts, and upscale only the final version to 1080p or 4K.
Video generation that preserves reference materials
The model also includes a feature that accepts images, short motion clips, and audio cues as references to improve consistency in people, products, and styles. According to the official documentation, it supports video generation using text, images, and reference materials; generation that connects the first and last frames; and video editing, extension, and upscaling. It can also generate speech, music, and sound effects. However, audio reference functionality is limited to an allowlist.
Availability and labeling
Gemini Omni 1.2 Flash is officially available through the Gemini API and Google Cloud, and is listed as being available across regions worldwide. Its coverage is also expanding to Google Vids and Gemini for business. Generated content includes an invisible SynthID watermark for identification, and provisioned throughput is available for enterprise environments.