Google DeepMind has released Gemini Omni 1.2 Flash, a video generation model now generally available across the Gemini API and Gemini for business. The headline numbers: 15-second native video clips (up from 10 seconds), a 30% price drop to $0.07 per second at 720p, and multi-turn conversational editing that lets users refine clips turn by turn rather than regenerating from scratch. The model also supports resolution drafting at 360p or 720p before upscaling to 1080p or 4K. The technical improvements center on reference adherence and consistency — the model's ability to hold faces, garments, brand typography, and color across camera angles and transitions. Google positions this as the unlock for commercial use cases: virtual try-on sequences, product catalog spins, branded video ads, and multi-character dialogue scenes with lip-sync. The examples shown include a 15-second virtual try-on with 360-degree turns, an animated exhibition poster built layer by layer, and a continuous FPV drone flythrough of an architectural space with zero jump cuts. The economics tell the real story. At $0.07 per second for 720p, a 15-second clip costs $1.05 to generate. The multi-turn editing workflow means a background swap or lighting change costs a fraction of full regeneration. Draft at 360p, iterate, then upscale — the cost curve rewards staying inside Google's pipeline. This is the classic platform playbook: reduce per-unit cost to drive volume, then capture the workflow. Google is deploying the model across an unusually wide surface area simultaneously: the Gemini API for developers, Google Cloud console for enterprises, Flow for scene direction, Google Vids for business users with Google Workspace subscriptions, the Gemini app for AI Ultra subscribers, and Google Ads for campaign scaling. Provisioned Throughput is available for enterprise-grade workloads, and all outputs carry SynthID watermarks. This breadth of deployment suggests Google views video generation not as a standalone product but as infrastructure woven into its existing business tools. The acknowledged limitations are revealing. Google notes ongoing development for "fast sports action, rapid script changes, and complex physics" — precisely the areas where current video generation models struggle most and where the gap between AI-generated and traditionally produced video remains widest. These are not small caveats; they define the boundary between AI video as a production tool and AI video as a replacement for production. For the broader market, this release accelerates the commoditization of short-form branded video. The combination of lower prices, longer clips, and multi-turn editing directly targets the social media ad production pipeline — a space currently served by agencies, freelancers, and tools like Canva. The 15-second clip length is not accidental; it maps precisely to the dominant social ad format. Google is not just selling a model — it is selling the elimination of a production workflow. The competitive dynamics are stark. This positions Google against Runway, Pika, and OpenAI's Sora in the API-accessible video generation market, but with a structural advantage: distribution through Google Ads, Workspace, and Cloud means the model reaches commercial buyers through channels they already use. The question is whether quality and consistency at these price points are sufficient for brand-sensitive commercial use, or whether the model becomes the high-volume, low-stakes tier while premium production holds its ground.
