Google opens Gemini Omni 1.2 Flash to all: videos up to 15 seconds and 30% lower pricing

Video generation was a lottery for a long time: press a button, get ten seconds, dislike the background—and pay to generate everything again. Google is trying to close precisely this gap. Gemini Omni 1.2 Flash is now available to everyone, and its main advantage is not clip length but how cheaply the result can be refined into the desired form.

What changed

Clips now run from 3 to 15 seconds, up from a previous ceiling of 10. The price of 720p has fallen by 30% to $0.07 per second. That means a 15-second video in 720p will cost about $1.05. The model is available through the Gemini API, Google AI Studio, Google Cloud Console, Flow, Google Vids, the Gemini app for AI Ultra subscribers, and Google Ads. Every video receives an invisible SynthID watermark.

That last point matters for products: if you integrate generation into your service, the origin of the content will be machine-readable automatically, removing some of the questions facing moderation and legal teams.

Conversational editing and cost savings

Google calls the main feature Conversational Editing. After receiving a video, you can type commands such as “make it evening instead of daytime” or “change the format to vertical.” The model edits the clip instead of generating a new one from scratch. According to Google, this works across the full 15-second duration.

That enables a cost-saving workflow. Create a draft in 360p or 720p, correct individual details with follow-up prompts, and upscale only the final version to 1080p or 4K. Previously, teams spent their budgets on expensive high-quality test runs. Now the costly pass is needed only once.

References and consistency

The model accepts character images, product photos, short motion references of up to three seconds, and audio prompts. Faces, clothing, packaging typography, and brand colors are expected to remain consistent as the camera angle changes. In Google's example, one photo of a model and images of three outfits produce a 15-second virtual fitting video with a 360-degree turn, preserving both the face and the cut of the clothing.

For e-commerce, this is a direct use case: product listings with video but no production crew. Google also promises fewer “melting” faces and deformations during camera movement, more natural dialogue with accurate lip synchronization, and differentiation between multiple speakers. Complex continuous camera movements, such as an FPV drone fly-through or a rapid pullback, should also contain fewer unwanted cuts.

Limitations and what to do

For now, all of this comes from the announcement. There are no independent tests of identity consistency or lip-sync performance in complex scenes, and the familiar word “promises” should be read literally. In addition, 4K and 1080p are billed separately, while the final cost depends on the number of editing iterations, whose pricing was not disclosed in the announcement.

The practical takeaway is simple. If you create advertising videos, virtual try-ons, or marketplace content, run the same task through a three-stage process: a low-resolution draft, conversational edits, and final upscaling. Calculate the actual cost of the finished video, not the price of generation time. That metric will show whether video generation has become a production tool or remains a toy.