Google has unveiled its upgraded Gemini Omni 1.2 Flash model specializing in AI video generation and processing. The new version introduces a set of technical improvements focused on boosting production efficiency and lowering costs, alongside enhanced control and interactive editing capabilities via sequential text and voice commands.
Longer duration and a 30% reduction in production costs
The new update increases the maximum length of a single clip from 10 seconds in the previous 1.1 version to 15 seconds, while giving users the freedom to customize display duration between 3 and 15 seconds in one-second increments. This move aims to enable content creators and marketers to produce social media ads and promotional product presentations in a single continuous clip without needing to merge separate footage.
Regarding pricing, Google has reduced processing costs by 30% compared to the previous version, setting the price for generating 720p video at $0.07 per second via the Gemini API and Google Cloud platform. The new mechanism allows developers to create and preview initial drafts at 360p or 720p to optimize consumption, and then upscale the final versions to 1080p or 4K upon final approval.
Conversational video editing and visual element consistency
The model provides advanced support for reference media, allowing it to be supplied with static product or character images, alongside reference motion clips of up to three seconds and guided audio tracks. This architecture helps maintain consistent facial features, brand colors, fonts, clothing details, and packaging specifics during camera angle changes and scene motion, serving virtual try-on applications for fashion and commercial products.
Gemini Omni 1.2 Flash also supports multi-turn conversational editing, eliminating the need to completely regenerate the video when modifications are desired. Users can now issue follow-up instructions to change elements such as:
- Replacing scene backgrounds or adjusting lighting and weather conditions.
- Swapping products within the frame while maintaining scene motion.
- Adjusting aspect ratios to fit various digital platforms.
- Improving lip-sync synchronization and generating natural interactive dialogues while isolating the voices of multiple speakers.
Availability and current technical limitations
The model is now available for direct use via Google AI Studio, the Gemini API, and Google Cloud, alongside its inclusion in Google enterprise tools such as Flow, Google Vids, and Google Ads, while the 15-second video generation feature has been allocated within the Gemini application for Google AI Ultra plan subscribers. All generated clips carry the invisible SynthID watermark to document the artificial source of the content.
On the other hand, Google acknowledged the existence of engineering limitations still under development in the current model, particularly when simulating fast athletic movements, sudden scene transitions, and phenomena relying on highly complex optical physics.