Google DeepMind Launches Gemini Omni 1.2 Flash with 15-Second Video Generation
Google DeepMind has released Gemini Omni 1.2 Flash, featuring a 30% price cut, 15-second native video generation, multi-turn conversational editing, and tighter reference adherence.
Google DeepMind has officially released Gemini Omni 1.2 Flash, marking its general availability across both the Gemini API and enterprise platforms. Announced by Product Manager Anish Nangia, the updated video generation model brings longer output windows, lowered generation costs, and improved reference consistency to help creators and businesses generate video assets at scale.
Cost Reductions and Extended Video Durations
A primary update in Gemini Omni 1.2 Flash is its revised pricing structure. Google DeepMind has reduced the cost of generation by 30% compared to version 1.1, bringing the price down to $0.07 per second for 720p resolution video. The model also introduces a flexible drafting and delivery pipeline designed to minimize compute expenses. Users can draft content at 360p or 720p resolution, refine specific details using targeted edits, and upscale finalized clips to 1080p or 4K.
In addition to lower pricing, the maximum native clip length has expanded from 10 seconds in version 1.1 to 15 seconds in version 1.2. Developers and creators can define output durations in precise one-second increments anywhere between 3 and 15 seconds. This added length allows single generations to cover complete social media ads, product explainer videos, and short narrative sequences without requiring artificial extensions.
Multi-Turn Conversational Editing and Reference Adherence
Gemini Omni 1.2 Flash introduces multi-turn conversational editing directly within the generation process. Rather than re-generating an entire sequence from scratch to fix a minor issue, users can issue conversational follow-up prompts to modify background settings, adjust ambient lighting, change weather conditions, swap products, or alter aspect ratios across full 15-second clips.
The model also demonstrates higher fidelity to input references. Users can guide video creation by supplying character sheets, product photos, style images, short motion reference clips lasting up to three seconds, and audio files. Omni 1.2 Flash maintains character faces, clothing drapes, brand typography, and specific color schemes across varying camera angles with minimal visual drift or morphing.
Camera control has also been upgraded to render continuous movements—such as sweeping first-person view (FPV) drone flights, wide orbits, and dramatic zooms—with fewer jump cuts. For scenes with spoken audio, the model improves lip-sync accuracy, dialogue fluidness, and multi-speaker conditioning by mapping sampled voice references to designated characters in a scene.
What it means for developers
For technical teams and software engineers, Gemini Omni 1.2 Flash provides enterprise-grade reliability backed by Google Cloud infrastructure. Generative outputs include invisible SynthID watermarking for tracking, and enterprise users can leverage Provisioned Throughput for consistent performance.
Developers can prototype applications in Google AI Studio or deploy production workloads through the Google Cloud console. For teams building multi-model software stacks, developers can try top AI models cheaply through one API at https://apixoai.online, simplifying cross-provider integrations alongside native Google cloud tools.
While Google DeepMind notes ongoing work to address limitations in fast sports motion, complex physical simulations, and rapid script adjustments, Omni 1.2 Flash is immediately accessible across multiple entry points. Beyond the direct developer API, the model is rolling out across Google Vids, Google Ads, the Flow environment, and the Gemini mobile app for Google AI Ultra subscribers.
Source: Gemini Omni 1.2 Flash: Fast and cost-effective with longer clips — Google DeepMind. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key
