The News, Explained

On November 5, Google made Gemini Omni 1.2 Flash, its video generation and editing model, generally available on the paid Gemini API. An API lets another program send requests to the model. The new version can make a clip from 3 to 15 seconds in one-second steps. The previous 1.1 model topped out at 10 seconds. Source Source Source

A reference clip is a short example of something you want the new video to follow, such as a camera move. In the API, a 3–10 second output can use up to three reference clips, while an 11–15 second output can use one. Each reference can be at most three seconds. To use two different camera moves as references, for example, choose an output of 10 seconds or less. Google’s announcement describes audio cues, but its current API guide says uploading audio inputs or references is unsupported. Source Source

For the same 720p resolution under Standard pricing, Google lists video output at about US$0.70 for 10 seconds and US$1.04 for 15 seconds. Billing uses output tokens, whose count depends on the length and resolution. Inputs such as text or images are billed separately. These examples cover one video output through the paid API; other Google apps and the overall cost of a project have separate considerations. Source

There is also a change for programs that already call 1.1. Google’s API guide says that starting December 3, requests sent to the existing 1.1 model code will automatically run on 1.2 at the new rates. A workflow can therefore start getting output from a different model version even if its request code stays the same. Source Source

OYOPICK’s Take

Choose the references you need before choosing the length. Staying at 10 seconds leaves more options if the shot has to follow several motion examples, while 15 seconds may help if you need a longer continuous scene. To compare production costs, consider how many times you might need to generate the same length and resolution. Google’s Standard examples are a useful starting point for estimating the cost of one output. Source Source