NederlandsGoogle releases Gemini Omni 1.2 Flash: 15-second AI videos, 30 per cent cheaper
Google's new video model makes clips of 15 seconds instead of 10 and is 30 per cent cheaper for developers. A clip can now be edited in conversation with the model. Google is still working on fast sports footage and complex physics.
Google DeepMind released Gemini Omni 1.2 Flash on 5 November, the newest version of its AI model that makes videos. A clip can now last 15 seconds in one go, compared with 10 seconds for version 1.1. The price also goes down: developers pay 30 per cent less, about 7 US cents per second of video in 720p quality. With version 1.1 it was about 10 US cents.
For the first time, Google calls an Omni model generally available. According to Google, companies can now build their regular work on it and reserve computing capacity in advance. All videos get an invisible watermark, SynthID, which lets software recognise that they were made with Google's AI.
First sketch, then finish
The length of a clip can be set per second, from 3 to 15 seconds. Anyone who wants to try things cheaply first makes a quick low-quality version: 360p costs about 2.3 US cents per second. The best clip can then be scaled up to 1080p (about 10.4 US cents per second) or 4K (about 20.9 US cents).
Editing now happens in conversation with the model. You ask, for example, for a different background, different lighting or a different format, and the model updates the same clip instead of starting again from scratch. According to Google, this works over the full 15 seconds. This makes it quick to create variants of a single advert. Developers can also extend a clip step by step, by 3 to 10 seconds each time, up to 30 seconds in total.
Less drifting and distortion
According to Google, the biggest gain is in holding on to details. In AI video, a face or a package can slowly change during a clip. Users can now steer the model with up to six example images of a character, product or style, with short moving fragments of up to three seconds, and with sound. These details are said to stay more consistent from the first frame to the last.
Google also mentions more natural conversations, lips that match the words better and separate control for each speaker. Long, smooth camera movements, such as a flight straight through a scene, are said to give far fewer unwanted jumps in the picture.
Google is open itself about what still needs to improve: fast sports footage, quick changes in the script and complex physics are still in development. All claims about the quality come from Google itself. Tests by independent researchers can show how big the step is in practice.
Where to use it
Developers and companies use the model through the Gemini API and Google Cloud. It is also in the video maker Flow and in Google Ads. In Google Vids, anyone with a Google account or a Workspace subscription can make clips of 15 seconds at no extra cost. In the Gemini app, it will become available to subscribers of Google AI Ultra. Anyone still using version 1.1 through the API will switch automatically to the new version from 3 December.
What does this mean for the road to the singularity?
Longer clips, a lower price and editing without starting again make AI video more usable for everyday work, such as advertising and media. Consistency is also becoming more important: if people and objects stay the same for longer, longer stories in pictures come closer. The next step lies with what Google itself already names: fast movement and physics that is correct.
What is the technological singularity? Read the explanation →
