Gemini Omni 1.2 Flash: Google's Video Generation Update

Gemini Omni 1.2 Flash, Google's model for AI video generation and editing, is now generally available on Google Cloud. The update extends clips to 15 seconds, improves reference adherence, and lowers Google's generation price. It is also already live on vega.chat as Gemini Omni and through api.vega.chat, available worldwide, including Russia, without a VPN or foreign account.

Gemini Omni 1.2 Flash: what changes in this update

The upgrade from Omni 1.1 Flash focuses on practical production needs: longer clips, more consistent subjects, and fewer reasons to restart a generation.

Generate and edit up to 15 seconds

The previous single-generation limit was 10 seconds. A 15-second AI video gives you room for a product demonstration, a short scene, or a complete ad rather than one fragment of it.

Lower pricing on Google Cloud

Google lists 720p generation at $0.07 per second, 30% less than Omni 1.1 Flash, and says other resolutions receive similar savings without giving exact figures in the announcement. These are Google Cloud prices, not VEGA pricing.

Better image and video reference adherence

The model follows supplied references more closely, so products, characters, and brand details stay consistent across shots. For a retailer, that means keeping the actual product recognizable instead of generating something merely in the same category.

Google also reports more natural dialogue, clearer lip-sync, and more distinct conditioning for multiple speakers. Audio references are supported subject to allowlisting, so that capability is not universally accessible yet.

Conversational video editing and 4K upscaling

Conversational video editing lets you refine an existing clip instead of generating everything again. The examples Google gives include replacing a product, changing the lighting, and adjusting the camera angle.

That changes the iteration process: keep the parts that work, then ask for a targeted revision. It is particularly useful when a scene is close to usable but one detail needs correcting.

The draft-to-final workflow separates exploration from delivery. You draft at 360p or 720p, pick the strongest result, then upscale it to 1080p or 4K. Every experiment no longer has to run as a final render.

The announcement also describes customers using video extension and video-to-video restyling. It does not give precise technical limits for those workflows, so the 15-second limit should not be read as the complete specification.

AI video ads, product clips, and customer examples

The clearest use cases start with material a business already has. Existing product photos can become product spins, lifestyle clips, virtual try-on videos, and AI video ads across a catalog. Reference consistency matters here: the product needs to survive changes in scene and presentation.

Customer examples in the announcement show how the update fits into real workflows:

• Artlist uses reference-based generation and video extension; FLORA uses video-to-video restyling.


• invideo generates full scenes in one prompt and praises the model's prompt adherence, physics, sound, and speed.


• Krea highlights character and scene replacements — targeted edits rather than complete restarts.


• Picsart iterates on drafts, then renders AI dramas in 4K for more than 100 million creators.


• Photoroom powers its Video Ad Maker and Product Animator with the model; its CEO says 15-second output was a top user request.


• QuickFrame AI produced a complete 15-second, 1080p connected-TV ad in a single generation.

These are customer-reported results, not a promise that every prompt yields a finished ad on the first try. They do show the direction of the update: longer, more controllable clips that need less assembly.

Try Gemini Omni 1.2 Flash on vega.chat

On vega.chat, choose the Gemini Omni video model. Version 1.2 is already available there for text-to-video, image-to-video, reference-to-video, and conversational editing.

A useful first test: start with a product image and a specific scene description, then refine the result conversationally. Check product details and motion before choosing a final output.

Developers can reach the model through the OpenRouter-compatible api.vega.chat to integrate video generation and editing into applications and agents.

On the Google Cloud side, the announcement adds Provisioned Throughput and imperceptible SynthID watermarking. Availability inside Gemini for business remains "coming soon" — separate from the model's current general availability and its existing access through VEGA.