Google has released Gemini Omni 1.2 Flash, an update to its AI video model that stretches clips from 10 to 15 seconds and cuts the price by 30%. It is also the first time Gemini Omni is generally available, which means companies can build it into production tools.
For developers, the headline number is $0.07 per second of 720p video. That is 30% below Omni 1.1. Google’s cost advice is to test ideas at low resolution first and upscale only the keeper, up to 4K.
Length is now set in one-second steps between 3 and 15 seconds. According to Google’s announcement, the model also edits through conversation: you can swap a product, change the weather or lighting, or switch the aspect ratio across a full 15-second clip, one request at a time, instead of generating it again.
Google also says the model follows references better. Inputs can be character, product or style pictures, audio hints, and motion clips no longer than three seconds, and faces, clothing, packaging text and brand colours are meant to stay the same from shot to shot. Dialogue is meant to sound more natural, with clearer lip-sync and separate voices for several speakers, and long camera moves such as FPV flythroughs and orbits should come with fewer unwanted jump cuts.
The model reaches people in several places. It is available through the Gemini API and Google Cloud, with Provisioned Throughput for heavier use, and Google Flow, Google Ads and Google Vids get it too. Google Vids users pay nothing extra for the 15-second clips. The Gemini app gets 15-second videos first for Google AI Ultra subscribers, as the rollout continues. Output carries Google’s invisible SynthID watermark.
Google’s own demos show what it has in mind: a 15-second virtual try-on clip built from one model photo and front-and-back shots of three outfits, and a single unbroken FPV flight through a piazza and up into the Pantheon’s dome. Another clip swaps the background turn by turn, from a tent to a forest to a castle, while the characters and camera framing stay put.
The first two clips Google highlights are a macro science explainer built from one style frame, with spelled-out labels that follow the motion, and an octopus replaced by a dancer in a second turn while the fluid movement carries over. Google adds that business customers and startups are already using the model.
Fast sports action and complicated physics remain weak spots, Google says, along with scripts that change quickly. The company’s demo clips are its own choices, so the quality across everyday prompts will only show once people use it. This follows the Genie 4 world model that Google DeepMind unveiled in late October.
